Daily Tech Digest - December 14, 2025


Quote for the day:

“It is never too late to be what you might have been.” -- George Eliot


Six questions to ask when crafting an AI enablement plan

As we near the end of 2025, there are two inconvenient truths about AI that every CISO needs to take into their heart. Truth #1: Every employee who can is using generative AI tools for their job. Even when your company doesn’t provide an account for them, even when your policy forbids it, even when the employee has to pay out of pocket. Truth #2: Every employee who uses generative AI will (or likely has already) provided this AI with internal and confidential company information. ... In the case of AI, this refers to the difference between the approved business apps that are trusted to access company data and the growing number of untrusted and unmanaged apps that have access to that data without the knowledge of IT or security teams. Essentially, employees are using unmonitored devices, which can hold any number of unknown AI apps, and each of those apps can introduce a whole lot of risk to sensitive corporate data. ... Simply put, organizations cannot afford to wait any longer to get a handle on AI governance. ... So now, the job is to craft an AI enablement plan that promotes productive use and throttles reckless behaviors. ... Think back to the mid‑2000s, when SaaS crept into the enterprise through expense reports and project trackers. IT tried to blacklist unvetted domains, finance balked at credit‑card sprawl, and legal wondered whether customer data belonged on “someone else’s computer.” Eventually, we accepted that the workplace had evolved, and SaaS became essential to modern business.


Why most enterprise AI coding pilots underperform (Hint: It's not the model)

When organizations introduce agentic tools without addressing workflow and environment, productivity can decline. A randomized control study this year showed that developers who used AI assistance in unchanged workflows completed tasks more slowly, largely due to verification, rework and confusion around intent. The lesson is straightforward: Autonomy without orchestration rarely yields efficiency. ... Security and governance, too, demand a shift in mindset. AI-generated code introduces new forms of risk: Unvetted dependencies, subtle license violations and undocumented modules that escape peer review. Mature teams are beginning to integrate agentic activity directly into their CI/CD pipelines, treating agents as autonomous contributors whose work must pass the same static analysis, audit logging and approval gates as any human developer. GitHub’s own documentation highlights this trajectory, positioning Copilot Agents not as replacements for engineers but as orchestrated participants in secure, reviewable workflows. ... Under the hood, agentic coding is less a tooling problem than a data problem. Every context snapshot, test iteration and code revision becomes a form of structured data that must be stored, indexed and reused. As these agents proliferate, enterprises will find themselves managing an entirely new data layer: One that captures not just what was built, but how it was reasoned about. 


Enabling small language models to solve complex reasoning tasks

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) developed a collaborative approach where an LLM does the planning, then divvies up the legwork of that strategy among smaller ones. Their method helps small LMs provide more accurate responses than leading LLMs like OpenAI’s GPT-4o, and approach the precision of top reasoning systems such as o1, while being more efficient than both. Their framework, called “Distributional Constraints by Inference Programming with Language Models” (or “DisCIPL”), has a large model steer smaller “follower” models toward precise responses when writing things like text blurbs, grocery lists with budgets, and travel itineraries. ... You may think that larger-scale LMs are “better” at complex prompts than smaller ones when it comes to accuracy and efficiency. DisCIPL suggests a surprising counterpoint for these tasks: If you can combine the strengths of smaller models instead, you may just see an efficiency bump with similar results. The researchers note that, in theory, you can plug in dozens of LMs to work together in the DisCIPL framework, regardless of size. In writing and reasoning experiments, they went with GPT-4o as their “planner LM,” which is one of the models that helps ChatGPT generate responses. 


Key trends accelerating Industrial Secure Remote Access (ISRA) Adoption

As essential maintenance and diagnostic activities continue to shift toward remote and digital execution, they become exposed to cyber risks that were not present when plants, fleets, and factories operated as isolated, closed systems. Compounding the challenge, many industrial organizations still lack the expertise and skill sets to select and operate the proper technologies that establish secure remote connections efficiently and securely. This, unfortunately, results in operational delays and slower response in critical or emergency situations. Industrial Cyber emphasizes that controlled, identity-bound, and fully auditable access to critical tasks is key to ensuring secure remote access functions as an operational and business enabler—without introducing new pathways for malicious actors. ... Compounding the risk, OT environments frequently rely on legacy hardware that lacks modern encryption capabilities, leaving these connections especially vulnerable. By centralizing access governance, securely managing vendor credentials, streamlining access-request workflows, and maintaining consistent audit trails, industrial organizations can regain control over third-party access. ... Industrial Cyber recognizes two solutions from SSH. 1) PrivX OT is purpose-built for industrial environments. The solution provides passwordless, keyless, and just-in-time industrial secure remote access using short-lived certificates and micro-segmentation to reduce risk. 2) NQX delivers quantum-safe, high-speed network encryption for site-to-site connectivity.


Navigating AI Liability: What Businesses That Utilize AI Need to Know

Cybercriminals can now use generative AI to create extremely convincing deepfakes. These deepfakes can then be used for corporate espionage, identity theft and phishing scams. AI software may end up automatically aggregating and analyzing huge amounts of data from multiple sources. This can increase privacy invasion risks when comprehensive profiles of people are compiled without their awareness or consent. AI systems which experience glitches or malfunctions, let others have unauthorized access to them, or lack robust security could lead to sensitive data being exposed. ... It is risky for your business to publish AI-generated content because AI models are trained on vast amounts of copyrighted material. The models thus end up not always creating original material, and sometimes create material which is identical to or extremely similar to copyrighted content. “It was the AI’s fault” will not be a valid argument in court if this happens to your business. Ignorance is not a defense in a copyright infringement claim. ... Content that is fully generated by AI has no copyright protection. AI-generated content that is significantly edited by humans may receive copyright protection, but the situation is murky. Original content that is created by humans and is then slightly edited or optimized by AI will usually receive full copyright protection. A lot of businesses now document the process of content creation to prove that humans created the content and preserve copyright protection.


When the Cloud Comes Home: What DBAs Need to Know About Cloud Repatriation

One of the main drivers for cloud repatriation is cost. Early cloud migrations were often justified by projected savings because there would be no more hardware to maintain. Furthermore, the cloud promised flexible scaling and pay-as-you-go pricing. Nevertheless, for many enterprises, those savings have proven elusive. Data-intensive workloads, in particular, can rack up significant cloud bills. Every I/O operation, network transfer, and storage request adds up. When workloads are steady and predictable, the cloud’s on-demand elasticity can actually become more expensive than on-prem capacity. DBAs, who often have a front-row seat to performance and utilization metrics, can play a crucial role in identifying when cloud costs are out of alignment with business value. ... In highly regulated industries, compliance concerns are another driver. Regulations such as HIPAA, PCI-DSS, GDPR and more, require your applications and the data they access to be secure and controlled. Organizations may find that managing sensitive data in the cloud introduces risk, especially when data residency, auditability, or encryption requirements evolve. Repatriating workloads can restore a sense of control and predictability—key traits valued by DBAs. ... Today’s computing needs demand an IT architecture that embraces the cloud, but also on premises workloads, including the mainframe. Remember, data gravity attracts applications to where the data resides. 


SaaS price hikes put CIOs’ budgets in a bind

Subscription prices from major SaaS vendors have risen sharply in recent months, putting many CIOs in a bind as they struggle to stay within their IT budgets. ... While inflation may have driven some cost increases in past months, rates have since stabilized, meaning there are other factors at play, Tucciarone says. Vendors are justifying subscription price hikes with frequent product repackaging schemes, consumption-based subscription models, regional pricing adjustments, and evolving generative AI offerings, he adds. “Vendors are rationalizing this as the cost of innovation and gen AI development,” he says. ... SaaS data platforms fall into a similar category as other mission-critical applications, Aymé adds, because the cost of moving an organization’s data can be prohibitively expensive, in addition to the price of a new SaaS tool. Kunal Agarwal, CEO and cofounder of data observability platform Unravel Data, also pointed to price increases for data-related SaaS tools. Data infrastructure costs, including cloud data warehouses, lakehouses, and analytics platforms, have risen 30% to 50% in the past year, he says. Several factors are driving cost increases, including the proliferation of computing-intensive gen AI workloads and a lack of visibility into organizational consumption, he adds. “Unlike traditional SaaS, where you’re paying for seats, these platforms bill based on consumption, making costs highly variable and difficult to predict,” Agarwal says.


How to simplify enterprise cybersecurity through effective identity management

“It is challenging for a lot of organizations to get a complete picture of what their assets are and what controls apply to those assets,” Persaud says. He explains that Deloitte’s identity solution assisted the customer in connecting users with the assets they utilized. As they discovered these assets, they were able to fine-tune the security controls that were applied to each in a more refined fashion. “If the system is going to [process] financial data and other private information, we need to put the right controls in place on the identity side,” he says. “We’ve been able to bring those two pieces together by correlating discovery of assets with discovery of identity and lining that up with controls from the IT asset management system.” ... “If you think from a broader risk management perspective, this has been fundamental to our security model,” he says. The ability to simply track the locations of employees and assign risk accordingly is a significant advancement in risk monitoring for a company growing its international presence. The company looks out for instances of impossible travel, such as if an employee has entered the system in one location and then in another at a distant location that they could not have possibly reached during a specified period, an alert is raised. Security analysts also use the software to scan for risky sign-ins. If a user logs in from an IP that has been blacklisted, an alert is raised. They have increasingly relied on conditional access policies that rely on monitoring user behavior. 



When an AI Agent Says ‘I Agree,’ Who’s Consenting?

The most autonomous agents can execute a chain of actions related to a transaction—such as comparing, booking, paying, forwarding the invoice. The broader the autonomy, the tighter the frame: precise contractual rules, allow-lists, budgets, a kill-switch, clear user notices, and, where required, electronic signatures. At this point the question stops being technical and becomes legal: under what framework does each agent-made click have effect, on whose authority, and with what safeguards? European law and national laws already offer solid anchors—agency and online contracting, signatures and secure payments, fair disclosure—now joined by the newer eIDAS 2 and the AI Act. ... Under European law, an AI agent has no will of its own. It is a means of expressing—or failing to express—someone’s will. Legally, someone always consents: the user (consumer) or a representative in the civil law sense. If an agent “accepts” an offer, we are back to agency: the act binds the principal only within the authority granted; beyond that, it is unenforceable. The agent is not a new subject of law. ... Who is on the hook if consent is tainted? First, the business that designs the onboarding. Europe’s Digital Services Act (DSA) bans deceptive interfaces (“dark patterns”) that materially impair a user’s ability to make a free, informed choice. A pushy interface can support a finding of civil fraud and a regulatory breach. Second, the principal is bound only within the mandate. 


AI cybercrime agents will strike in 2026: Are defenders ready?

The prediction itself isn’t novel. What’s sobering is the math behind it—and the widening gap between how fast organisations can defend versus how quickly they’re being attacked. “The groups that convert intelligence into monetisation the fastest will set the tempo,” Rashish Pandey, VP of marketing & communications for APAC at Fortinet, told journalists at a media briefing earlier this week. “Throughput defines impact.” This isn’t about whether AI will be weaponised—that’s already happening. The urgent question is whether defenders can close what Fortinet calls the “tempo differential” before autonomous AI agents fundamentally alter the economics of cybercrime. ... The evolution extends beyond speed. Fortinet’s predictions highlight how attackers are weaponising generative AI for rapid data analysis—sifting through stolen information to identify the most valuable targets and optimal extortion strategies before defenders even detect the breach. This aligns with broader attack trends: ransomware operations increasingly blend system disruption with data theft and multi-stage extortion. Critical infrastructure sectors—healthcare, manufacturing, utilities—face heightened risk as operational technology systems become targets. ... “The ‘skills gap’ is less about scarcity and more about alignment—matching expertise to the reality of machine-speed, data-driven operations,” Pandey noted during the briefing.

No comments:

Post a Comment