Daily Tech Digest - May 31, 2024

Flawed AI Tools Create Worries for Private LLMs, Chatbots

The research underscores that the rush to integrate AI into business processes does pose risks, especially for companies that are giving LLMs and other generative-AI applications access to large repositories of data. ... The risks posed by the adoption of next-gen artificial intelligence and machine learning (AI/ML) are not necessarily due to the models, which tend to have smaller attack surfaces, but the software components and tools for developing AI applications and interfaces, says Dan McInerney, lead AI threat researcher with Protect AI, an AI application security firm. "There's not a lot of magical incantations that you can send to an LLM and have it spit out passwords and sensitive info," he says. "But there's a lot of vulnerabilities in the servers that are used to host LLMs. The [LLM] is really not where you're going to get hacked — you're going to get hacked from all the tools you use around the LLM." ... "Exploitation of this vulnerability could affect the immediate functioning of the model and can have long-lasting effects on its credibility and the security of the systems that rely on it," Synopsys stated in its advisory. 


Cyber resiliency is a key focus for us: Balaji Rao, Area VP – India & SAARC, Commvault

Referring to the classical MITRE framework, the recommendation is to “shift right” – moving focus towards recovery. After thoroughly assessing risks and implementing various tools, it’s crucial to have a solid recovery plan in place. Customers are increasingly concerned about scenarios where both their primary and disaster recovery (DR) systems are compromised by ransomware, and their backups are unavailable. According to a Microsoft report, in 98% of successful ransomware cases, backups are disabled. To address this concern, the strategy involves building a cyber resilient framework that prioritises recovery. ... For us, AI serves multiple purposes, primarily enhancing efficiency, scanning for threats, and addressing customer training and enablement needs. From a security perspective, we leverage AI extensively to detect ransomware-related risks. Its rapid data processing capabilities allow for thorough scanning across vast datasets, enabling pattern matching and identifying changes indicative of potential threats. We’ve integrated AI into our threat scanning solutions, strengthening our ability to detect and mitigate malware by leveraging comprehensive malware databases.


The importance of developing second-line leaders

Developing second-line leaders helps your business unit or function succeed at a whole new level: When your teams know that leadership development is a priority, they start preparing for future roles. The top talent will cultivate their skills and equip themselves for leadership positions, enhancing overall team performance. As the cascading effect builds, this proactive development has a multiplicative impact, especially if competition within the team remains healthy. It's also important for your personal growth as a leader: The most fulfilling aspect is the impact on yourself. Measuring your leadership success by contribution, attribution, and legacy, developing capable successors fulfils all three criteria. It ensures you contribute effectively, gain recognition for building strong teams, and leave a lasting legacy through the leaders you've developed. ... It starts with the self. Begin with delegation without abdication or evasion of accountability. This skill is a cornerstone of effective leadership, involving the entrusting of responsibilities to others while empowering them to assume ownership and make informed decisions.


Navigating The AI Revolution: Balancing Risks And Opportunities

Effective trust management requires specific approaches, such as robust monitoring systems, rigorous auditing processes and well-defined incident response plans. More importantly, in order for any initiative to address AI risks to be successful, we as an industry need to build a workforce of trained professionals. Those operating in the digital trust domain, including cybersecurity, privacy, assurance, risk and governance of digital technology, need to understand AI before building controls around it. The ISACA AI survey revealed that 85% of digital trust professionals say they will need to increase their AI skills and knowledge within two years to advance or retain their jobs. This highlights the importance of continuous learning and adaptation for cybersecurity professionals in the era of AI. Gaining a deeper understanding of how AI-powered attacks are altering the threat landscape, along with how AI can be effectively utilized by security practitioners, will be essential. As security professionals learn more about AI, they need to ensure that the methods being deployed align with an enterprise’s overarching need to maintain trust with its stakeholders.


CISO‘s Guide to 5G Security: Risks, Resilience and Fortifications

A strong security posture requires granular visibility into 5G traffic and automated security enforcement to effectively thwart attackers, protect critical services, and safeguard against potential threats to assets and the environment. This includes a focus on detecting and preventing attacks at all layers, interface and threat vector — from equipment (PEI) and subscriber (SUPI) identification, applications, signaling, data, network slices, malware, ransomware and more. ... To accomplish the task at hand brought about by 5G, CISOs must be prepared to provide a swift response to known and unknown threats in real time with advanced AI and machine learning, automation and orchestration tools. As connotation shifts from viewing 4G as a more consumer-focused mobile network to the power of private 5G when embedded across enterprise infrastructure, any kind of lateral network movement can bring about damage. ... Strategy and solution start with zero trust and can go as far as an entire 5G SOC dedicated to the nuances brought about by the next-gen network. The change and progress 5G promises is only as significant as our ability to protect networks and infrastructure from malicious actors, threats, and attacks.


Cloud access security brokers (CASBs): What to know before you buy

CASBs sit between an organization’s endpoints and cloud resources, acting as a gateway that monitors everything that goes in or out, providing visibility into what users are doing in the cloud, enforcing access control policies, and looking out for security threats. ... The original use case for CASBs was to address shadow IT. When security execs deployed their first CASB tools, they were surprised to discover how many employees had their own personal cloud storage accounts, where they squirreled away corporate data. CASB tools can help security teams discover and monitor unauthorized or unmanaged cloud services being used by employees. ... Buying a CASB tool can be complex. There’s a laundry list of possible features that fall within the broad CASB definition (DLP, SWG, etc.) And CASB tools themselves are part of a larger trend toward SSE and SASE platforms that include features such as ZTNA or SD-WAN. Enterprises need to identify their specific pain points — whether that’s regulatory compliance or shadow IT — and select a vendor that meets their immediate needs and can also grow with the enterprise over time.


What is model quantization? Smaller, faster LLMs

Why do we need quantization? The current large language models (LLMs) are enormous. The best models need to run on a cluster of server-class GPUs; gone are the days where you could run a state-of-the-art model locally on one GPU and get quick results. Quantization not only makes it possible to run a LLM on a single GPU, it allows you to run it on a CPU or on an edge device. ... As you might expect, accuracy may be an issue when you quantize a model. You can evaluate the accuracy of a quantized model against the original model, and decide whether the quantized model is sufficiently accurate for your purposes. For example, TensorFlow Lite offers three executables for checking the accuracy of quantized models. You might also consider MQBench, a benchmark and framework for evaluating quantization algorithms under real-world hardware deployments that uses PyTorch. If the degradation in accuracy from post-training quantization is too high, then one alternative is to use quantization aware training.


Europe Declares War on Tech Spoofing

In the new Payment Services Regulation, members of the European Parliament argued that messaging services such as WhatsApp, digital platforms such as Facebook, or marketplaces such as Amazon and eBay could be liable for scams that originate on their platforms, on a par with banks and other payment service providers. ... Europe’s new payment regulations are now up for negotiation in Brussels. Large US tech firms and messaging apps are pushing to lower the liability risk. They argue banks, not them, should be responsible. With spoofing or impersonation scams, the fraudulent transaction occurs on banking service portals, not the platforms. And so, banks themselves should enhance their security measures or pay the price. Banks, not surprisingly, disagree. They cannot control the entry points that fraudsters use to reach consumers, whether it is by phone, messaging apps, online ads, or the dark web. Why shouldn’t telecom network operators, messaging, and other digital platforms also be obliged to avoid fraudsters from reaching consumers and if they fail, be held liable?


Process mining helps IT leaders modernize business operations

Process mining provides the potential to enable organizations make quicker, more informed decisions when overhauling business processes by leveraging data for insights. By using the information gleaned from process mining, companies can better streamline workflows, enhance resource allocation, and automate repetitive tasks. ... Successful deployment and maintenance of process mining requires a clear vision from the management team and board, Mortello says, as well as commitment and persistence. “Process mining doesn’t usually yield immediate, tangible results, but it can offer unique insights into how a company operates,” he says. “A leadership team with a long-term vision is crucial to ensure the technology is utilized to its full potential.” It’s also important to thoroughly analyze processes prior to “fixing” them. “Make sure you have a good handle on the process you think you have and the ones you really have,” Constellation Research’s Wang says. “What we see across the board is a quick realization that what’s assumed and what’s done is very different.”


Could the Next War Begin in Cyberspace?

In a cyberwar, disinformation campaigns will likely be used to spread misinformation and collect data that can be leveraged to sway public opinion on key issues, Janzen says. "We can build very sophisticated security systems, but so long as we have people using those systems, they will be targeted to willingly or unwillingly allow malicious actors into those systems." ... How long a cyberspace war might last is inherently unpredictable, characterized by its persistent and ongoing nature, Menon says. "In contrast to conventional wars, marked by distinct start and end points, cyber conflicts lack geographical constraints," he notes. "These battles involve continuous attacks, defenses, and counterattacks." The core of cyberspace warfare lies in understanding algorithms, devising methods to breach them, and inventing new technologies to dismantle legacy systems, Menon says. "These factors, coupled with the relatively low financial investment required, contribute to the sporadic and unpredictable nature of cyberwars, making it challenging to anticipate when they may commence."



Quote for the day:

"It's fine to celebrate success but it is more important to heed the lessons of failure." -- Bill Gates

No comments:

Post a Comment