Daily Tech Digest: prompt engineering

Showing posts with label prompt engineering. Show all posts

Daily Tech Digest - September 05, 2025

Quote for the day:

"Little minds are tamed and subdued by misfortune; but great minds rise above it." -- Washington Irving

Understanding Context Engineering: Principles, Practices, and Its Distinction from Prompt Engineering

Context engineering is the strategic design, management, and delivery of relevant information—or “context”—to AI systems in order to guide, constrain, or enhance their behavior. Unlike prompt engineering, which primarily focuses on crafting effective input prompts to direct model outputs, context engineering involves curating, structuring, and governing the broader pool of information that surrounds and informs the AI’s decision-making process. In practice, context engineering requires an understanding of not only what the AI should know at a given moment but also how information should be prioritized, retrieved, and presented. It encompasses everything from assembling relevant documents and dialogue history to establishing policies for data inclusion and exclusion. ... While there is some overlap between the two domains, context engineering and prompt engineering serve distinct purposes and employ different methodologies. Prompt engineering is concerned with the formulation of the specific text—the “prompt”—that is provided to the model as an immediate input. It is about phrasing questions, instructions, or commands in a way that elicits the desired behavior or output from the AI. Successful prompt engineering involves experimenting with wording, structure, and sometimes even formatting to maximize the performance of the language model on a given task.

How AI and Blockchain Are Transforming Tenant Verification in India

While artificial intelligence provides both intelligence and speed, Blockchain technology provides the essential foundation of trust and security. Blockchain functions as a permanent digital record – meaning that once information is set, it can’t be changed or deleted by third parties. This feature is particularly groundbreaking for ensuring a safe and clear rental history. Picture this: the rental payments and lease contracts of your tenants could all be documented as ‘smart contracts’ using Blockchain technology. ... The combination of AI and Blockchain signifies a groundbreaking transformation, enabling tenants to create ‘self-sovereign identities’ on the Blockchain — digital wallets that hold their verified credentials, which they fully control. When searching for rental properties, tenants can conveniently provide prospective landlords with access to certain details about themselves, such as their history of timely payments and police records. AI leverages secure and authentic Blockchain data to produce an immediate risk score for landlords to assess, ensuring a quick and reliable evaluation. This cohesive approach guarantees that AI outcomes are both rapid and trustworthy, while the decentralized nature of Blockchain safeguards tenant privacy by removing the necessity for central databases that may become susceptible over time.

Adversarial AI is coming for your applications

New research from Cato Networks threat intelligence report, revealed how threat actors can use a large language model jailbreak technique, known as an immersive world attack, to get AI to create infostealer malware for them: a threat intelligence researcher with absolutely no malware coding experience managed to jailbreak multiple large language models and get the AI to create a fully functional, highly dangerous, password infostealer to compromise sensitive information from the Google Chrome web browser. The end result was malicious code that successfully extracted credentials from the Google Chrome password manager. Companies that create LLMs are trying to put up guardrails, but clearly GenAI can make malware creation that much easier. AI-generated malware, including polymorphic malware, essentially makes signature-based detections nearly obsolete. Enterprises must be prepared to protect against hundreds, if not thousands, of malware variants. ... Enterprises can increase their protection by embedding security directly into applications at the build stage: this involves investing in embedded security that is mapped to OWASP controls; such as RASP, advanced Whitebox cryptography, and granular threat intelligence. IDC research shows that organizations protecting mobile apps often lack a solution to test them efficiently and effectively.

Top Pitfalls to Avoid When Responding to Cyber Disaster

Moving too quickly following an attack can also prompt staff to respond to an intrusion without first fully understanding the type of ransomware that was used. Not all ransomware is created equal and knowing if you were a victim of locker ransomware, double extortion, ransomware-as-a-service, or another kind of attack can make all the difference in how to respond because the goal of the attacker is different for each. ... The first couple hours after a ransomware incident is identified are critical. In those immediate hours, work quickly to identify and isolate affected systems and disconnect compromised devices from the network to prevent the ransomware from spreading further. Don’t forget to also preserve forensic evidence as you go, such as screenshots, relevant logs, anything to inform future law enforcement investigations or legal action. Once that has been done, notify the key stakeholders and the cyber insurance provider. ... After the dust settles, analyze how the attack was able to occur and put in place fixes to keep it from happening again. Identify the initial access point and method, and map how the threat actor moved through the network. What barriers were they able to move past, and which held them back? Are there areas where more segmentation is needed to reduce the attack surface? Do any security workflows or policies need to be modified?

How to reclaim control over your online shopping data

“While companies often admit to sharing user data with third parties, it’s nearly impossible to track every recipient. That lack of control creates real vulnerabilities in data privacy management. Very few organizations thoroughly vet their third-party data-sharing practices, which raises accountability concerns and increases the risk of breaches,” said Ian Cohen, CEO of LOKKER. The criminal marketplace for stolen data has exploded in recent years. In 2024, over 6.8 million accounts were listed for sale, and by early 2025, nearly 2.5 million stolen accounts were available at one point. ... Even limited purchase information can prove valuable to criminals. A breach exposing high-value transactions, for example, may suggest a buyer’s financial status or lifestyle. When combined with leaked addresses, that data can help criminals identify and target individuals more precisely, whether for fraud, identity theft, or even physical theft. ... One key mechanism is the right to be forgotten, a legal principle allowing individuals to request the removal of their personal data from online platforms. The European Union’s GDPR is the strongest example of this principle in action. While not as comprehensive as the GDPR, the US has some privacy protections, such as the California Consumer Privacy Act (CCPA), which allow residents to access or delete their personal data.

Mind the Gap: Agentic AI and the Risks of Autonomy

The ink is barely dry on generative AI and AI agents, and now we have a new next big thing: agentic AI. Sounds impressive. By the time this article comes out, there’s a good chance that agentic AI will be in the rear-view mirror and we’ll all be chasing after the next new big thing. Anyone for autonomous generative agentic AI agent bots? ... Some things on the surface seem more irresponsible than others, but for some, agentic AI apparently not so much. Debugging large language models, AI agents, and agentic AI, as well as implementing guardrails are topics for another time, but it’s important to recognize that companies are handing over those car keys. Willingly. Enthusiastically. Would you put that eighth grader in charge of your marketing department? Of autonomously creating collateral that goes out to your customers without checking it first? Of course not. ... We want AI agents and agentic AI to make decisions, but we must be intentional about the decisions they are allowed to make. What are the stakes personally, professionally, or for the organization? What is the potential liability when something goes wrong? And something will go wrong. Something that you never considered going wrong will go wrong. And maybe think about the importance of the training data. Isn’t that what we say when an actual person does something wrong? “They weren’t adequately trained.” Same thing here.

How software engineers and team leaders can excel with artificial intelligence

As long as software development and AI designers continue to fall prey to the substitution myth, we’ll continue to develop systems and tools that, instead of supposedly making humans lives easier/better, will require unexpected new skills and interventions from humans that weren’t factored into the system/tool design ... Software development covers a lot of ground, from understanding requirements, architecting, designing, coding, writing tests, code review, debugging, building new skills and knowledge, and more. AI has now reached a point where it can automate or speed up almost every part of the process. This is an exciting time to be a builder. A lot of the routine, repetitive, and frankly boring parts of the job, the "cognitive grunt work", can now be handled by AI. Developers especially appreciate the help in areas like generating test cases, reviewing code, and writing documentation. When those tasks are off our plate, we can spend more time on the things that really add value: solving complex problems, designing great systems, thinking strategically, and growing our skills. ... The elephant in the room is "whether AI will take over my job one day?". Until this year, I always thought no, but the recent technological advancements and new product offerings in this space are beginning to change my mind. The reality is that we should be prepared for AI to change the software development role as we know it.

6 browser-based attacks all security teams should be ready for in 2025

Phishing tooling and infrastructure has evolved a lot in the past decade, while the changes to business IT means there are both many more vectors for phishing attack delivery, and apps and identities to target. Attackers can deliver links over instant messenger apps, social media, SMS, malicious ads, and using in-app messenger functionality, as well as sending emails directly from SaaS services to bypass email-based checks. Likewise, there are now hundreds of apps per enterprise to target, with varying levels of account security configuration. ... Like modern credential and session phishing, links to malicious pages are distributed over various delivery channels and using a variety of lures, including impersonating CAPTCHA, Cloudflare Turnstile, simulating an error loading a webpage, and many more. The variance in lure, and differences between different versions of the same lure, can make it difficult to fingerprint and detect based on visual elements alone. ... Preventing malicious OAuth grants being authorized requires tight in-app management of user permissions and tenant security settings. This is no mean feat when considering the 100s of apps in use across the modern enterprise, many of which are not centrally managed by IT and security teams

JSON Config File Leaks Azure ActiveDirectory Credentials

"The critical risk lies in the fact that this file was publicly accessible over the Internet," according to the post. "This means anyone — from opportunistic bots to advanced threat actors — could harvest the credentials and immediately leverage them for cloud account compromise, data theft, or further intrusion." ... To exploit the flaw, an attacker can first use the leaked ClientId and ClientSecret to authenticate against Azure AD using the OAuth2 Client Credentials flow to acquire an access token. Once this is acquired, the attacker then can send a GET request to the Microsoft Graph API to enumerate users within the tenant. This allows them to collect usernames and emails; build a list for password spraying or phishing; and/or identify naming conventions and internal accounts, according to the post. The attacker also can query the Microsoft Graph API to enumerate OAuth2 permission grants within the tenant, revealing which applications have been authorized and what scopes, or permissions, they hold. Finally, the acquired token allows an attacker to use group information to identify privilege clusters and business-critical teams, thus exposing organizational structure and identifying key targets for compromise, according to the post. ... "What appears to be a harmless JSON configuration file can in reality act as a master key to an organization’s cloud kingdom," according to the post.

Data centers are key to decarbonizing tech’s AI-fuelled supply chain

Data center owners and operators are uniquely positioned to step up and play a larger, more proactive role in this by pushing back on tech manufacturers in terms of the patchy emissions data they provide, while also facilitating sustainable circular IT product lifecycle management/disposal solutions for their users and customers. ... The hard truth, however, is that any data center striving to meet its own decarbonization goals and obligations cannot do so singlehandedly. It’s largely beholden to the supply chain stakeholders upstream. At the same time, their customers/users tend to accept ever shortening usage periods as the norm. Often, they overlook the benefits of achieving greater product longevity and optimal cost of ownership through the implementation of product maintenance, refurbishment, and reuse programmes. ... As a focal point for the enablement of the digital economy, data centers are ideally placed to take a much more active role: by lobbying manufacturers, educating users and customers about the necessity and benefits of changing conventional linear practices in favour of circular IT lifecycle management and recycling solutions. Such an approach will not only help decarbonize data centers themselves but the entire tech industry supply chain – by reducing emissions.

Daily Tech Digest - August 06, 2025

Quote for the day:

"What you do has far greater impact than what you say." -- Stephen Covey

“Man in the Prompt”: New Class of Prompt Injection Attacks Pairs With Malicious Browser Extensions to Issue Secret Commands to LLMs

The so-called “Man in the Prompt” attack presents two priority risks. One is to internal LLMs that store sensitive company data and personal information, in the belief that it is appropriately fenced off from other software and apps. The other risk comes from particular LLMs that are broadly integrated into workspaces, such as Google Gemini’s interaction with Google Workspace tools such as Mail and Docs. This category of prompt injection attacks applies not just to any type of browser extension, but any model or deployment of LLM. And the malicious extension requires no special permissions to work, given that the DOM access already provides everything it needs. ... The other proof-of-concept targets Google Gemini, and by extension any elements of Google Workspace it has been integrated with. Gemini is meant to automate routine and tedious tasks in Workspace such as email responses, document editing and updating contacts. The trouble is that it has almost complete access to the contents of these accounts as well as anything the user has access permission for or has had shared with them by someone else. Prompt injection attacks conducted by these extensions can not only steal the contents of emails and documents with ease, but complex queries can be fed to the LLM to target particular types of data and file extensions; the autocomplete function can also be abused to enumerate available files.

EU seeks more age verification transparency amid contentious debate

The EU is considering setting minimum requirements for online platforms to disclose their use of age verification or age estimation tools in their terms and conditions. The obligation is contained in a new compromise draft text of the EU’s proposed law on detecting and removing online child sex abuse material (CSAM), dated July 24 and seen by MLex. A discussion of the proposal, which contains few other changes to a previous draft, is scheduled for September 12. The text also calls for online platforms to perform mandatory scans for CSAM, which critics say could result in false positives and break end-to-end cryptography. ... The way age verification is set to work under the OSA is described as a “privacy nightmare” by PC Gamer, but the article stands in stark contrast to the vague posturing of the political class. Author Jacob Ridley acknowledges the possibility for double-blind methods of age assurance among those that do not require any personal information at all to be shared with the website or app the individual is trying to access. At the same time, many age verification systems do not work this way. Also, age assurance pop-ups can be spoofed, and those spoofs could harvest a wealth of valuable personal information Privado ID Co-founder Evan McMullen calls it “like using a sledgehammer to crack a walnut.” McMullen, of course, prefers a decentralized approach that leans on zero-knowledge proofs (ZKPs).

AI Is Changing the Cybersecurity Game in Ways Both Big and Small

“People are rushing now to get [MCP] functionality while overlooking the security aspect,” he said. “But once the functionality is established and the whole concept of MCP becomes the norm, I would assume that security researchers will go in and essentially update and fix those security issues over time. But it will take a couple of years, and while that is taking time, I would advise you to run MCP somehow securely so that you know what’s going on.” Beyond the tactical security issues around MCP, there are bigger issues that are more strategic, more systemic in nature. They involve the big changes that large language models (LLMs) are having on the cybersecurity business and the things that organizations will have to do to protect themselves from AI-powered attacks in the future ... The sheer volume of threat data, some of which may be AI generated, demands more AI to be able to parse it and understand it, Sharma said. “It’s not humanly possible to do it by a SOC engineer or a vulnerability engineer or a threat engineer,” he said. Tuskira essentially functions as an AI-powered security analyst to detect traditional threats on IT systems as well as threats posed to AI-powered systems. Instead of using commercial AI models, Sharma adopted open-source foundation models running in private data centers. Developing AI tools to counter AI-powered security threats demands custom models, a lot of fine-tuning, and a data fabric that can maintain context of particular threats, he said.

AI burnout: A new challenge for CIOs

To take advantage of the benefits of smart tools and avoid overburdening the workforce, the board of directors must carefully manage their deployment. “As leaders, we must set clear limits, encourage training without overwhelming others, and open spaces for conversation about how people are experiencing this transition,” Blázquez says. “Technology must be an ally, not a threat, and the role of leadership will be key in that balance.” “It is recommended that companies take the first step. They must act from a preventative, humane, and structural perspective,” says De la Hoz. “In addition to all the human, ethical, and responsible components, it is in the company’s economic interest to maintain a happy, safe, and mission-focused workforce.” Regarding increasing personal productivity, he emphasizes the importance of “valuing their efforts, whether through higher salary returns or other forms of compensation.” ... From here, action must be taken, “implementing contingency plans to alleviate these areas.” One way: working groups, where the problems and barriers associated with technology can be analyzed. “From here, use these KPIs to change my strategy. Or to set it up, because often what happens is that I deploy the technology and forget how to get that technology adopted.”

CIOs need a military mindset

While the battlefield feels very far away from the boardroom, this principle is something that CIOs can take on board when they’re tasked with steering a complex digital programme. Step back and clear the path so that you can trust your people to deliver; that’s when the real progress gets made. Contrary to popular belief, the military is not rigidly hierarchical. In fact, it teaches individuals to operate with autonomy within defined parameters. Officers set the boundaries of a mission and step back, allowing you to take full ownership of your actions. This approach is supported by the OODA Loop, a framework that cultivates awareness and decisive action under pressure. ... Resilience is perhaps the hardest leadership trait to teach and the most vital to embody. Military officers are taught to plan exhaustively, train rigorously, and prepare for all scenarios, but they’re also taught that ‘the first casualty of war is the plan.’ Adaptability under pressure is a non-negotiable mindset for you to adopt and instil in your team. When your team feels supported to grow, they stop fearing change and start responding to it; it is here that adaptability and resilience become second nature. There is also a practical opportunity to bring these principles in-house, as veterans transitioning out of the army may bring with them a refreshed leadership approach. Because they’re often confident under pressure and focused on outcomes, their transferrable skills allow them to thrive in the corporate world.

Backend FinOps: Engineering Cost-Efficient Microservices in the Cloud

Integrating cost management directly into Infrastructure-as-Code (IaC) frameworks such as Terraform enforces fiscal responsibility at the resource provisioning phase. By explicitly defining resource constraints and mandatory tagging, teams can preemptively mitigate orphaned cloud expenditures. ... Integrating cost awareness directly within Continuous Integration and Delivery (CI/CD) pipelines ensures proactive management of cloud expenditures throughout the development lifecycle. Tools such as Infracost automate the calculation of incremental cloud costs introduced by individual code changes. ... Cost-based pre-merge testing frameworks reinforce fiscal prudence by simulating peak-load scenarios prior to code integration. Automated tests measured critical metrics, including ninety-fifth percentile response times and estimated cost per ten thousand requests, to ensure compliance with established financial performance benchmarks. Pull requests failing predefined cost-efficiency criteria were systematically blocked. ... Comprehensive cost observability tools such as Datadog Cost Dashboards combine billing metrics with Application Performance Monitoring (APM) data, directly supporting operational and cost-related SLO compliance.

5 hard truths of a career in cybersecurity — and how to navigate them

Leadership and HR teams often gatekeep by focusing exclusively on candidates with certain educational degrees or specific credentials, typically from vendors such as Cisco, Juniper, or Palo Alto. Although Morrato finds this somewhat understandable given the high cost of hiring in cybersecurity, he believes this approach unfairly filters out capable individuals who, in a different era, would have had more opportunities. ... Because most team managers elevate from technical roles, they often lack the leadership and interpersonal skills needed to foster healthy team cultures or manage stakeholder relationships effectively. This cultural disconnect has a tangible impact on individuals. “People who work in security functions don’t always feel safe — psychologically safe — doing so,” Budge explains. ... Cybersecurity teams must also rethink how they approach risk, as relying solely on strict, one-size-fits-all controls is no longer tenable, Mistry says. Instead, he advocates for a more adaptive, business-aligned framework that considers overall exposure rather than just technical vulnerabilities. “Can I live with this risk? Can I not live with this risk? Can I do something to reduce the risk? Can I offload the risk? And it’s a risk conversation, not a ‘speeds and feeds’ conversation,” he says, emphasizing that cybersecurity leaders must actively build relationships across the organization to make these conversations possible.

How AI amplifies these other tech trends that matter most to business in 2025

Agentic AI is an artificial intelligence system capable of independently planning and executing complex, multistep tasks. Built on foundation models, these agents can autonomously perform actions, communicate with one another, and adapt to new information. Significant advancements have emerged, from general agent platforms to specialized agents designed for deep research. ... Application-specific semiconductors are purpose-built chips optimized to perform specialized tasks. Unlike general-purpose semiconductors, they are engineered to handle specific workloads (such as large-scale AI training and inference tasks) while optimizing performance characteristics, including offering superior speed, energy efficiency, and performance. ... Cloud and edge computing involve distributing workloads across locations, from hyperscale remote data centers to regional hubs and local nodes. This approach optimizes performance by addressing factors such as latency, data transfer costs, data sovereignty, and data security. ... Quantum-based technologies use the unique properties of quantum mechanics to execute certain complex calculations exponentially faster than classical computers; secure communication networks; and produce sensors with higher sensitivity levels than their classical counterparts.

Differentiable Economics: Strategic Behavior, Mechanisms, and Machine Learning

Differential economics is related to but different from the recent progress in building agents that achieve super-human performance in combinatorial games such as chess and Go. First, economic games such as auctions, oligopoly competition, or contests typically have a continuous action space expressed in money, and opponents are modeled as draws from a prior distribution that has continuous support. Second, differentiable economics is focused on modeling and achieving equilibrium behavior. The second opportunity in differentiable economics is to use data-driven methods and machine learning to discover rules, constraints, and affordances—mechanisms—for economic environments that promote good outcomes in the equilibrium behavior of a system. Mechanism design solves the inverse problem of game theory, finding rules of strategic interaction such that agents in equilibrium will effect an outcome with desired properties. Where possible, mechanisms promote strong equilibrium solution concepts such as dominant strategy equilibria, making it strategically easy for agents to participate. Think of a series of bilateral negotiations between buyers and a seller that is replaced by an efficient auction mechanism with simple dominant strategies for agents to report their preferences truthfully.

Ownership Mindset Drives Innovation: Milwaukee Tool CEO

“Empowerment was not a free-for-all,” Richman explained. In fact, the company recently changed the wording around its core values from “empowerment” to “extreme ownership” to reflect the importance of accountability for results. Emphasizing ownership can also help employees do what is best for the company as a whole rather than just their own teams, particularly when it comes to reallocating resources. ... Surprises and setbacks are an unavoidable cost of trying new things while innovating. Since organizations cannot avoid these issues, leaders and employees need to discuss them frankly and quickly enough to minimize the downside while seizing the upside. “[Being] candid is the most challenging cultural element of any company,” Richman said. “And we believe that it really leads to success or failure.” … In successful cultures, teams, people, parts of the organization can bring problems up and bring them up in a way to be able to say, ‘How are we going to rally the troops as one team, come together, fix it, and figure out why we got into this mess, and what are we going to do to not do it again?’” Candor is a two-way street. To build trust, leaders need to provide an honest assessment of the state of the company and the path forward — a “candid communication of where you are,” Richman said.

Daily Tech Digest - March 16, 2025

Quote for the day:

"Absolute identity with one's cause is the first and great condition of successful leadership." -- Woodrow Wilson

What Do You Get When You Hire a Ransomware Negotiator?

Despite calls from law enforcement agencies and some lawmakers urging victims not to make any ransom payment, the demand for experienced ransomware negotiators remains high. The negotiators say they provide a valuable service, even if the victim has no intention to pay. They bring skills into an incident that aren't usually found in the executive suite - strategies for dealing with criminals. ... Negotiation is more a thinking game, in which you try to outsmart the hackers to buy time and ascertain valuable insight, said Richard Bird, a ransomware negotiator who draws much of his skills from his past stint as a law enforcement crises aversion expert - talking people out of attempting suicide or negotiating with kidnappers for the release of hostages. "The biggest difference is that when you are doing a face-to-face negotiation, you can pick-up lots of information from a person on their non-verbal communications such as eye gestures, body movements, but when you are talking to someone over email or messaging apps that can cause some issues - because you have got to work out how the person might perceive," Bird said. One advantage of online negotiation is that it gives the negotiator time to reflect on what to tell the hackers.

Managing Data Security and Privacy Risks in Enterprise AI

While enterprise AI presents opportunities to achieve business goals in a way not previously conceived, one should also understand and mitigate potential risks associated with its development and use. Even AI tools designed with the most robust security protocols may still present a multitude of risks. These risks include intellectual property theft, privacy concerns when training data and/or output data may contain personally identifiable information (PII) or protected health information (PHI), and security vulnerabilities stemming from data breaches and data tampering. ... Privacy and data security in the context of AI are interdependent disciplines that often require simultaneous consideration and action. To begin with, advanced enterprise AI tools are trained on prodigious amounts of data processed using algorithms that should be—but are not always—designed to comply with privacy and security laws and regulations. ... Emerging laws and regulations related to AI are thematically consistent in their emphasis on accountability, fairness, transparency, accuracy, privacy, and security. These principles can serve as guideposts when developing AI governance action plans that can make your organization more resilient as advances in AI technology continue to outpace the law.

Mastering Prompt Engineering with Functional Testing: A Systematic Guide to Reliable LLM

OutputsCreating efficient prompts for large language models often starts as a simple task… but it doesn’t always stay that way. Initially, following basic best practices seems sufficient: adopt the persona of a specialist, write clear instructions, require a specific response format, and include a few relevant examples. But as requirements multiply, contradictions emerge, and even minor modifications can introduce unexpected failures. What was working perfectly in one prompt version suddenly breaks in another. ... What might seem like a minor modification can unexpectedly impact other aspects of a prompt. This is not only true when adding a new rule but also when adding more detail to an existing rule, like changing the order of the set of instructions or even simply rewording it. These minor modifications can unintentionally change the way the model interprets and prioritizes the set of instructions. The more details you add to a prompt, the greater the risk of unintended side effects. By trying to give too many details to every aspect of your task, you increase as well the risk of getting unexpected or deformed results. It is, therefore, essential to find the right balance between clarity and a high level of specification to maximise the relevance and consistency of the response.

You need to prepare for post-quantum cryptography now. Here’s why

"In some respects, we're already too late," said Russ Housley, founder of Vigil Security LLC, in a panel discussion at the conference. Housley and other speakers at the conference brought up the lesson from the SHA-1 to SHA-2 hashing-algorithm transition, which began in 2005 and was supposed to take five years but took about 12 to complete — "and that was a fairly simple transition," Housley noted. In a different panel discussion, InfoSec Global Vice President of Cryptographic Research & Development Vladimir Soukharev called the upcoming move to post-quantum cryptography a "much more complicated transition than we've ever seen in cryptographic history." ... The asymmetric algorithms that NIST is phasing out are thought to be vulnerable to this. The new ones that NIST is introducing use even more complicated math that quantum computers probably can't crack (yet). Today, an attacker could watch you log into Amazon and capture the asymmetrically-encrypted exchange of the symmetric key that secures your shopping session. But that would be pointless because the attacker couldn't decrypt that key exchange. In five or 10 years, it'll be a different story. The attacker will be able to decrypt the key exchange and then use that stolen key to reveal your shopping session

Network Forensics: A Short Guide to Digital Evidence Recovery from Computer Networks

At a technical level, this discipline operates across multiple layers of the OSI model. At the lower layers, it examines MAC addresses, VLAN tags, and frame metadata, while at the network and transport layers, it analyses IP addresses, routing information, port usage, and TCP/UDP session characteristics. ... Network communications contain rich metadata in their headers—the “envelope” information surrounding actual content. This includes IP headers with source/destination addresses, fragmentation flags, and TTL values; TCP/UDP headers containing port numbers, sequence numbers, window sizes, and flags; and application protocol headers with HTTP methods, DNS query types, and SMTP commands. This metadata remains valuable even when content is encrypted, revealing communication patterns, timing relationships, and protocol behaviors. ... Encryption presents perhaps the most significant technical challenge for modern network forensics, with over 95% of web traffic now encrypted using TLS. Despite encryption, substantial metadata remains visible, including connection details, TLS handshake parameters, certificate information, and packet sizing and timing patterns. This observable data still provides significant forensic value when properly analyzed.

Modernising Enterprise Architecture: Bridging Legacy Systems with Jargon

The growing gap between enterprise-wide architecture and the actual work being done on the ground leads to manual processes, poor integration, and limits how effectively teams can work across modern DevOps environments — ultimately creating the next generation of rigid, hard-to-maintain systems — repeating the mistakes of the past. ... Instead of treating enterprise architecture as a walled-off function, Jargon enables continuous integration between high-level architecture and real-world software design — bridging the gap between enterprise-wide planning and hands-on development while automating validation and collaboration. ... Jargon is already working with organisations to bridge the gap between modern API-first design and legacy enterprise tooling, enabling teams to modernise workflows without abandoning existing systems. While our support for OpenAPI and JSON Schema is already in place, we’re planning to add XMI support to bring Jargon’s benefits to a wider audience of enterprises who use legacy architecture tools. By supporting XMI, Jargon will allow enterprises to unlock their existing architecture investments while seamlessly integrating API-driven workflows. This helps address the challenge of top-down governance conflicting with bottom-up development needs, enabling smoother collaboration across teams.

CAIOs are stepping out from the CIO’s shadow

The CAIO position as such is still finding its prime location in the org chart, Fernández says, often assuming a position of medium-high responsibility in reporting to the CDO and thus, in turn, to the CIO. “These positions that are being created are very ‘business partner’ style,” he says, “to make these types of products understood, what needs they have, and to carry them out.” Casado adds: “For me, the CIO does not have such a ‘business case’ component — of impact on the profit and loss account. The role of artificial intelligence is very closely tied to generating efficiencies on an ongoing basis,” as well as implying “continuous adoption.” “It is essential that there is this adoption and that implies being very close to the people,” he says. ... Garnacho agrees, stating that, in less mature AI development environments, the CIO can assume CAIO functions. “But as the complexity and scope of AI grows, the specialization of the CAIO makes the difference,” he says. This is because “although the CIO plays a fundamental role in technological infrastructure and data management, AI and its challenges require specific leadership. In our view, the CIO lays the technological foundations, but it is the CAIO who drives the vision.” In this emerging division of functions, other positions may be impacted by the emergence of the AI chief.

Forget About Cloud Computing. On-Premises Is All the Rage Again

Cloud costs have a tendency to balloon over time: Storage costs per GB of data might seem low, but when you’re dealing with terabytes of data—which even we as a three-person startup are already doing—costs add up very quickly. Add to this retrieval and egress fees, and you’re faced with a bill you cannot unsee. Steep retrieval and egress fees only serve one thing: Cloud providers want to incentivize you to keep as much data as possible on the platform, so they can make money off every operation. If you download data from the cloud, it will cost you inordinate amounts of money. Variable costs based on CPU and GPU usage often spike during high-performance workloads. A report by CNCF found that almost half of Kubernetes adopters found that they’d exceeded their budget as a result. Kubernetes is an open-source container orchestration software that is often used for cloud deployments. The pay-per-use model of the cloud has its advantages, but billing becomes unpredictable as a result. Costs can then explode during usage spikes. Cloud add-ons for security, monitoring, and data analytics also come at a premium, which often increases costs further. As a result, many IT leaders have started migrating back to on-premises servers. A 2023 survey by Uptime found that 33% of respondents had repatriated at least some production applications in the past year.

IT leaders are driving a new cloud computing era

CIOs have become increasingly frustrated with vendor pricing models that lock them into unpredictable and often unfavorable long-term commitments. Many find that mounting operational costs frequently outweigh the promised savings from cloud computing. It’s no wonder that leadership teams are beginning to shift gears, discussing alternative solutions that might better serve their best interests. ... Regional or sovereign clouds offer significant advantages, including compliance with local data regulations that ensure data sovereignty while meeting industry standards. They reduce latency by placing data centers nearer to users, enhancing service performance. Security is also bolstered, as these clouds can apply customized protection measures against specific threats. Additionally, regional clouds provide customized services that cater to local needs and industries and offer more responsive customer support than larger global providers. ... The pushback against traditional cloud providers is not driven only by unexpected costs; it also reflects enterprise demand for greater autonomy, flexibility, and a skillfully managed approach to technology infrastructure. Effectively navigating the complexities of cloud computing will require organizations to reassess their dependencies and stay vigilant in seeking solutions that align with their growth strategies.

How Intelligent Continuous Security Enables True End-to-End Security

Intelligent Continuous Security (TM) (ICS) is the next evolution — harnessing AI-driven automation, real-time threat detection and continuous compliance enforcement to eliminate these inefficiencies. ICS extends beyond DevSecOps to also close security gaps with SecOps, ensuring end-to-end continuous security across the entire software lifecycle. This article explores how ICS enables true DevOps transformation by addressing the shortcomings of traditional security, reducing friction across teams, and accelerating secure software delivery. ... As indicated in the article The Next Generation of Security “The Future of Security is Continuous. Security isn’t a destination — it’s a continuous process of learning, adapting and evolving. As threats become smarter, faster, and more unpredictable, security must follow suit.” Traditional security practices were designed for a slower, waterfall-style development process. ... Intelligent Continuous Security (ICS) builds on DevSecOps principles but goes further by embedding AI-driven security automation throughout the SDLC. ICS creates a seamless security layer that integrates with DevOps pipelines, reducing the friction that has long plagued DevSecOps initiatives. ... ICS shifts security testing left by embedding automated security checks at every stage of development.

Daily Tech Digest - February 25, 2025

Quote for the day:

"Empowerment is the magic wand that turns a frog into a prince. Never estimate the power of the people, through true empowerment great leaders are born." -- Lama S. Bowen

Service as Software Changes Everything

Service as software, also referred to as SaaS 2.0, goes beyond layering AI atop existing applications. It centers on the concept of automating business processes through intelligent APIs and autonomous services. The framework aims to eliminate human input and involvement through AI agents that act and react to conditions based on events, behavioral changes, and feedback. The result is autonomous software. “Traditional SaaS provides cloud-based tools where staff still do the work. Service as software flips that script. Instead of having staff do the work, you're making calls to an API or using software that does the work for you,” says Mark Strefford, founder of TimelapseAI, a UK-based consulting firm. ... CIOs and IT leaders should start small and iterate, experts say. As an organization gains confidence and trust, it can expand the autonomy of a SaaS 2.0 component. “More AI initiatives have failed from starting too big than too small,” Strefford notes. Consequently, it’s critical to understand the entire workflow, build in oversight and protections, establish measurement and validation tools, and stay focused on outcomes. A few factors can make or break an initiative, Giron says. Data quality and the ability to integrate across systems is crucial. A framework for standardization is critical. This includes cleaning, standardizing, and preparing legacy data.

The Missing Sustainability Perspective in Cloud Architecture

The Well-Architected Framework provides a structured approach to making architectural decisions. While it originally focused on operational, security, and financial trade-offs, the Sustainability Pillar introduces specific guidance for designing cloud solutions with minimal environmental impact. One key architectural trade-off is between performance efficiency and sustainability. While performance efficiency emphasizes speed and low latency, these benefits often come at the cost of over-provisioning resources. A more sustainable approach involves optimizing compute resources to ensure they are only consumed when necessary. Serverless computing solutions, such as AWS Lambda or Azure Functions, help minimize idle capacity by executing workloads only when triggered. Similarly, auto-scaling for containerized applications, such as Kubernetes Horizontal Pod Autoscaler (HPA) or AWS Fargate, ensures that resources are dynamically adjusted based on demand, preventing unnecessary energy consumption. Another critical balance is between cost optimization and sustainability. Traditional cost optimization strategies focus on reducing expenses, but without considering sustainability, businesses might make short-term cost-saving decisions that lead to long-term environmental inefficiencies. For example, many organizations store large volumes of data without assessing its relevance, leading to excessive storage-related energy use.

Quantum Computing Has Arrived; We Need To Prepare For Its Impact

Many now believe that the power and speed of quantum computing will enable us to address some of the biggest and most difficult problems our civilization faces. Problem-solving will be made possible by quantum computing’s unprecedented processing speed and predictive analytics. That is a remarkable near-term potential. Mckinsey & Company forecasts that Quantum Technologies could create an economic value in the market of up to $2 trillion by 2035. Quantum measuring and sensing is one field where quantum technologies have already made their appearance. Navigational devices and magnetic resonance imaging already employ it. Quantum sensors detect and quantify minute changes in time, gravity, temperature, pressure, rotation, acceleration, frequency, and magnetic and electric fields using the smallest amounts of matter and energy. Quantum will have a direct impact on many scientific fields, including biology, chemistry, physics, and mathematics. Industry applications will have an impact on a wide range of fields, including healthcare, banking, communications, commerce, cybersecurity, energy, and space exploration. In other words, any sector in which data is a component. More specifically, quantum technology has incredible potential to transform a wide range of fields, including materials science, lasers, biotechnology, communications, genetic sequencing, and real-time data analytics.

Industrial System Cyberattacks Surge as OT Stays Vulnerable

"There's a higher propensity for manufacturing organizations to have cloud connectivity just as a way of doing business, because of the benefits of the public cloud for manufacturing, like for predictive analytics, just-in-time inventory management, and things along those lines," he says, pointing to Transportation Security Administration rules governing pipelines and logistics networks as one reason for the difference. "There is purposeful regulation to separate the IT-OT boundary — you tend to see multiple kinds of ring-fence layers of controls. ... There's a more conservative approach to outside-the-plant connectivity within logistics and transportation and natural resources," Geyer says. ... When it comes to cyber-defense, companies with operational technology should focus on protecting their most important functions, and that can vary by organization. One food-and-beverage company, for example, focuses on the most important production zones in the company, testing for weak and default passwords, checking for the existence of clear-text communications, and scanning for hard-coded credentials, says Claroty's Geyer. "The most important zone in each of their plants is milk receiving — if milk receiving fails, everything else is critical path and nothing can work throughout the plant," he says.

How to create an effective incident response plan

“When you talk about BIA and RTOs [recovery time objective], you shouldn’t be just checking boxes,” Ennamli says. “You’re creating a map that shows you, and your decision-makers, exactly where to focus efforts when things go wrong. Basically, the nervous system of your business.” ... “And when the rubber hits the road during an actual incident, precious time is wasted on less important assets while critical business functions remain offline and not bringing in revenue,” he says. ... It’s vital to have robust communication protocols, says Jason Wingate, CEO at Emerald Ocean, a provider of brand development services. “You’re going to want a clear chain of command and communication,” he says. “Without established protocols, you’re about as effective as trying to coordinate a fire response with smoke signals.” The severity of the incident should inform the communications strategy, says David Taylor, a managing director at global consulting firm Protiviti. While cybersecurity team members actively responding to an incident will be in close contact and collaborating during an event, he says, others are likely not as plugged in or consistently informed. “Based on the assigned severity, stemming from the initial triage or a change to the level of severity based on new information during the response, governance should dictate the type, audience, and cadence of communications,” Taylor says.

AI-Powered DevOps: Transforming CI/CD Pipelines for Intelligent Automation

Traditional software testing faces challenges as organizations must assess codes to ensure they do not downgrade system performance or introduce bugs. Applications with extensive functionalities are time-consuming as they demand several test cases. They must ensure appropriate management, detailing their needs and advancing critical results in every scope. Nonetheless, smoke and regression testing ensures the same test cases are conducted, leading to time-consuming activities. The difficulty makes it hard for the traditional approach to have critical coverage of what is needed, and it is challenging to ensure that every approach can be tackled appropriately, channeling value toward the demanded selection. ... Using ML-driven test automation leads to increased efficiency in managing repetitive tasks. These automated measures ensure an accelerated testing approach, allowing teams to work with better activities. ML also integrates quality assessment into the software, marking an increasingly beneficial way to attend to individual requirements to ensure every software is assessed for high risk, potential failures and critical functions, which achieve a better post-deployment result. Additionally, using ML automation leads to cost savings, enabling testing cycles to have minimal operational costs as they are automated and prevent defects from being deployed within the software.

Prompt Engineering: Challenges, Strengths, and Its Place in Software Development's Future

Prompt engineering and programming share the goal of instructing machines but differ fundamentally in their methodologies. While programming relies on formalized syntax, deterministic execution, and precision to ensure consistency and reliability, prompt engineering leverages the adaptability of natural language. This flexibility, however, introduces certain challenges, such as ambiguity, variability, and unpredictability. ... Mastering prompt engineering requires a level of knowledge and expertise comparable to programming. While it leverages natural language, its effective use demands a deep understanding of AI model behavior, the application of specific techniques, and a commitment to continuous learning. Similar to programming, prompt engineering involves continual learning to stay proficient with a variety of evolving techniques. A recent literature review by OpenAI and Microsoft analyzed over 1,500 prompt engineering-related papers, categorizing the various strategies into a formal taxonomy. This literature review is indicative of the continuous evolution of prompt engineering, requiring practitioners to stay informed and refine their approaches to remain effective.

Avoiding vendor lock-in when using managed cloud security services

An ideal managed cloud security provider should take an agnostic approach. Their solution should be compatible with whatever CNAPP or CSPM solution you use. This gives you maximum flexibility to find the right provider without locking yourself into a specific solution. Advanced services may even enable you to take open-sourced tooling and get to a good place before expanding to a full cloud security solution. You could also partner with a managed cloud security service that leverages open standards and protocols. This approach will allow you to integrate new or additional vendors while reducing your dependency on proprietary technology. Training and building in-house knowledge also helps. A confident service won’t keep their knowledge to themselves and helps enable and provide training to your team along the way. ... And there’s IAM—a more complex but equally concerning component of cloud security. In recent news, a few breaches started with low-level credentials being obtained before the attackers self-escalated themselves to gain access to sensitive information. This is often due to overly permissive access given to humans and machines. It’s also one of the least understood components of the cloud. Still, if your managed cloud security service truly understands the cloud, it won’t ignore IAM, the foundation of cloud security.

Observability Can Get Expensive. Here’s How to Trim Costs

“At its core, the ‘store it all’ approach is meant to ensure that when something goes wrong, teams have access to everything so they can pinpoint the exact location of the failure in their infrastructure,” she said. “However, this has become increasingly infeasible as infrastructure becomes more complex and ephemeral; there is now just too much to collect without massive expense.” ... “Something that would otherwise take developers weeks to do — take an inventory of all telemetry collected and eliminate the lower value parts — can be available at the click of a button,” she said. A proper observability platform can continually analyze telemetry data in order to have the most up-to-date picture of what is useful rather than a one-time, manual audit “that’s essentially stale as soon as it gets done,” Villa said. “It’s less about organizations wanting to pay less for observability tools, but they’re thinking more long-term about their investment and choosing platforms that will save them down the line,” she said. “The more they save on data collection, the more they can reinvest into other areas of observability, including new signals like profiling that they might not have explored yet.” Moving from a “store it all” to a “store intelligently” strategy is not only the future of cost optimization, Villa said, but can also help make the haystack of data smaller

The Aftermath of a Data Breach

For organizations, the aftermath of a data breach can be highly devastating. In an interconnected world, a single data vulnerability can cascade into decades of irreversible loss – intellectual, monetary, and reputational. The consequences paralyze even the most established businesses, uprooting them from their foundation. ... The severity of a data breach often depends on how long it goes undetected; however, identifying the breach is where the story actually begins. From containing the destruction and informing authorities to answering customers and paying for their damages, the road to business recovery is long and grueling. ... Organizations must create, implement, and integrate a data management policy in their organizational setup that provides a robust framework for managing data throughout its entire lifecycle, from creation to disposal. This policy should also include a data destruction policy that specifies data destruction methods, data wiping tools, type of erasure verification, and records of data destruction. It should further cover media control and sanitization, incident reporting, the roles and responsibilities of the CIO, CISO, and privacy officer. Using a professional software-based data destruction tool erases data permanently from IT assets including laptops, PCs, and Mac devices.

Daily Tech Digest - February 11, 2025

Quote for the day:

"Your worth consists in what you are and not in what you have." -- Thomas Edison

Protecting Your Software Supply Chain: Assessing the Risks Before Deployment

Given the vast number of third-party components used in modern IT, it's unrealistic to scrutinize every software package equally. Instead, security teams should prioritize their efforts based on business impact and attack surface exposure. High-privilege applications that frequently communicate with external services should undergo product security testing, while lower-risk applications can be assessed through automated or less resource-intensive methods. Whether done before deployment or as a retrospective analysis, a structured approach to PST ensures that organizations focus on securing the most critical assets first while maintaining overall system integrity. ... While Product Security Testing will never prevent a breach of a third party out of your control, it is necessary to allow organizations to make informed decisions about their defensive posture and response strategy. Many organizations follow a standard process of identifying a need, selecting a product, and deploying it without a deep security evaluation. This lack of scrutiny can leave them scrambling to determine the impact when a supply chain attack occurs. By incorporating PST into the decision-making process, security teams gain critical documentation, including dependency mapping, threat models, and specific mitigations tailored to the technology in use.

Google’s latest genAI shift is a reminder to IT leaders — never trust vendor policy

Entities out there doing things you don’t like are always going to be able to get generative AI (genAI) services and tools from somebody. You think large terrorist cells can’t use their money to pay somebody to craft LLMs for them? Even the most powerful enterprises can’t stop it from happening. But, that may not be the point. Walmart, ExxonMobil, Amazon, Chase, Hilton, Pfizer and Toyota and the rest of those heavy-hitters merely want to pick and choose where their monies are spent. Big enterprises can’t stop AI from being used to do things they don’t like, but they can make sure none of it is being funded with their money. If they add a clause to every RFP that they will only work with model-makers that agree to not do X, Y, or Z, that will get a lot of attention. The contract would have to be realistic, though. It might say, for instance, “If the model-maker later chooses to accept payments for the above-described prohibited acts, they must reimburse all of the dollars we have already paid and must also give us 18 months notice so that we can replace the vendor with a company that will respect the terms of our contracts.” From the perspective of Google, along with Microsoft, OpenAI, IBM, AWS and others, the idea is to take enterprise dollars on top of government contracts.

Is Fine-Tuning or Prompt Engineering the Right Approach for AI?

It’s not just about having access to GPUs — it’s about getting the most out of proprietary data with new tools that make fine-tuning easier. Here’s why fine-tuning is gaining traction:Better results with proprietary data: Fine-tuning allows businesses to train models on their own data, making the AI much more accurate and relevant to their specific tasks. This leads to better outcomes and real business value. Easier than ever before: Tools like Hugging Face’s Open Source libraries, PyTorch and TensorFlow, along with cloud services, have made fine-tuning more accessible. These frameworks simplify the process, even for teams without deep AI expertise. Improved infrastructure: The rising availability of powerful GPUs and cloud-based solutions has made it much easier to set up and run fine-tuning at scale. While fine-tuning opens the door to more customized AI, it does require careful planning and the right infrastructure to succeed. ... As enterprises accelerate their AI adoption, choosing between prompt engineering and fine-tuning will have a significant impact on their success. While prompt engineering provides a quick, cost-effective solution for general tasks, fine-tuning unlocks the full potential of AI, enabling superior performance on proprietary data.

Shifting left without slowing down

On the one hand, automation enabled by GenAI tools in software development is driving unprecedented developer productivity, further emphasizing the gap created by manual application security controls, like security reviews or threat modeling. But in parallel, recent advancements in code understanding enabled by these technologies, together with programmatic policy-as-code security policies, enable a giant leap in the value security automation can bring. ... The first step is recognizing security as a shared responsibility across the organization, not just a specialized function. Equipping teams with automated tools and clear processes helps integrate security into everyday workflows. Establishing measurable goals and metrics to track progress can also provide direction and accountability. Building cross-functional collaboration between security and development teams sets the foundation for long-term success. ... A common pitfall is treating security as an afterthought, leading to disruptions that strain teams and delay releases. Conversely, overburdening developers with security responsibilities without proper support can lead to frustration and neglect of critical tasks. Failure to adopt automation or align security goals with development objectives often results in inefficiency and poor outcomes.

How To Approach API Security Amid Increasing Automated Attack Sophistication

We’ve now gone from ‘dumb’ attacks—for example, web-based attacks focused on extracting data from third parties and on a specific or single vulnerability—to ‘smart’ AI-driven attacks often involving picking an actual target, resulting in a more focused attack. Going after a particular organization, perhaps a large organization or even a nation-state, instead of looking for vulnerable people is a significant shift. The sophistication is increasing as attackers manipulate request payloads to trick the backend system into an action. ... Another element of API security is being aware of sensitive data. Personal Identifiable Information (PII) is moving through APIs constantly and is vulnerable to theft or data exfiltration. Organizations do not often pay attention to vulnerabilities. Still, they pay attention when the result is damage to their organization through leaked PII, stolen finances, or brand reputation. ... The security teams know the network systems and the infrastructure well but don't understand the application behaviors. The DevOps team tends to own the applications but doesn’t see anything in production. This split boundary in most organizations makes it ripe for exploitation. Many data exfiltration cases fall in this no man’s land since an authenticated user executes most incidents.

Top 5 ways attackers use generative AI to exploit your systems

Gen AI tools help criminals pull together different sources of data to enrich their campaigns — whether this is group social profiling, or targeted information gleaned from social media. “AI can be used to quickly learn what types of emails are being rejected or opened, and in turn modify its approach to increase phishing success rate,” Mindgard’s Garraghan explains. ... The traditionally difficult task of analyzing systems for vulnerabilities and developing exploits can be simplified through use of gen AI technologies. “Instead of a black hat hacker spending the time to probe and perform reconnaissance against a system perimeter, an AI agent can be tasked to do this automatically,” Mingard’s Garraghan says. ... “This sharp decrease strongly indicates that a major technological advancement — likely GenAI — is enabling threat actors to exploit vulnerabilities at unprecedented speeds,” ReliaQuest writes. ... Check Point Research explains: “While ChatGPT has invested substantially in anti-abuse provisions over the last two years, these newer models appear to offer little resistance to misuse, thereby attracting a surge of interest from different levels of attackers, especially the low skilled ones — individuals who exploit existing scripts or tools without a deep understanding of the underlying technology.”

Why firewalls and VPNs give you a false sense of security

VPNs and firewalls play a crucial role in extending networks, but they also come with risks. By connecting more users, devices, locations, and clouds, they inadvertently expand the attack surface with public IP addresses. This expansion allows users to work remotely from anywhere with an internet connection, further stretching the network’s reach. Moreover, the rise of IoT devices has led to a surge in Wi-Fi access points within this extended network. Even seemingly innocuous devices like Wi-Fi-connected espresso machines, meant for a quick post-lunch pick-me-up, contribute to the proliferation of new attack vectors that cybercriminals can exploit. ... More doesn’t mean better when it comes to firewalls and VPNs. Expanding a perimeter-based security architecture rooted in firewalls and VPNs means more deployments, more overhead costs, and more time wasted for IT teams – but less security and less peace of mind. Pain also comes in the form of degraded user experience and satisfaction with VPN technology for the entire organization due to backhauling traffic. Other challenges like the cost and complexity of patch management, security updates, software upgrades, and constantly refreshing aging equipment as an organization grows are enough to exhaust even the largest and most efficient IT teams.

Building Trust in AI: Security and Risks in Highly Regulated Industries

AI hallucinations have emerged as a critical problem, with systems generating plausible but incorrect information - for instance, AI fabricated software dependencies, such as PyTorture, leading to potential security risks. Hackers could exploit these hallucinations by creating malicious components masquerading as real ones. In another case, an AI libelously fabricated an embezzlement claim, resulting in legal action - marking the first time AI was sued for libel. Security remains a pressing concern, particularly with plugins and software supply chains. A ChatGPT plugin once exposed sensitive data due to a flaw in its OAuth mechanism, and incidents like PyTorch’s vulnerable release over Christmas demonstrate the risks of system exploitation. Supply chain vulnerabilities affect all technologies, while AI-specific threats like prompt injection allow attackers to manipulate outputs or access sensitive prompts, as seen in Google Gemini. ... Organizations can enhance their security strategies by utilizing frameworks like Google’s Secure AI Framework (SAIF). These frameworks highlight security principles, including access control, detection and response systems, defense mechanisms, and risk-aware processes tailored to meet specific business needs.

When LLMs become influencers

Our ability to influence LLMs is seriously circumscribed. Perhaps if you’re the owner of the LLM and associated tool, you can exert outsized influence on its output. For example, AWS should be able to train Amazon Q to answer questions, etc., related to AWS services. There’s an open question as to whether Q would be “biased” toward AWS services, but that’s almost a secondary concern. Maybe it steers a developer toward Amazon ElastiCache and away from Redis, simply by virtue of having more and better documentation and information to offer a developer. The primary concern is ensuring these tools have enough good training data so they don’t lead developers astray. ... Well, one option is simply to publish benchmarks. The LLM vendors will ultimately have to improve their output or developers will turn to other tools that consistently yield better results. If you’re an open source project, commercial vendor, or someone else that increasingly relies on LLMs as knowledge intermediaries, you should regularly publish results that showcase those LLMs that do well and those that don’t. Benchmarking can help move the industry forward. By extension, if you’re a developer who increasingly relies on coding assistants like GitHub Copilot or Amazon Q, be vocal about your experiences, both positive and negative.

Deepfakes: How Deep Can They Go?

Metaphorically, spotting deepfakes is like playing the world’s most challenging game of “spot the difference.” The fakes have become so sophisticated that the inconsistencies are often nearly invisible, especially to the untrained eye. It requires constant vigilance and the ability to question the authenticity of audiovisual content, even when it looks or sounds completely convincing. Recognizing threats and taking decisive actions are crucial for mitigating the effects of an attack. Establishing well-defined policies, reporting channels, and response workflows in advance is imperative. Think of it like a citywide defense system responding to incoming missiles. Early warning radars (monitoring) are necessary to detect the threat; anti-missile batteries (AI scanning) are needed to neutralize it; and emergency services (incident response) are essential to quickly handle any impacts. Each layer works in concert to mitigate harm. ... If a deepfake attack succeeds, organizations should immediately notify stakeholders of the fake content, issue corrective statements, and coordinate efforts to remove the offending content. They should also investigate the source, implement additional verification measures, and provide updates to rebuild trust and consider legal action.

Pages