The local computing rebellion
Users are increasingly frustrated with the mandatory subscriptions, constant online requirements, and bundled artificial intelligence features pushed by major software companies. This dissatisfaction is driving a "local computing rebellion," where individuals and businesses are returning to software that runs independently on their own devices and stores data locally. This movement isn't a complete rejection of cloud services, which remain valuable for large-scale operations and collaboration, but rather a demand for choice and control over personal computing environments. People are tired of hidden dependencies—like authentication or telemetry—that can cause "local" apps to break during internet outages or when subscriptions lapse. In response, businesses are advised to audit their software dependencies, establish clear data storage policies, and prioritize flexibility in procurement, ensuring they can choose offline, local deployment options. To remain competitive, cloud providers must stop forcing constant connectivity and instead earn customer trust by offering transparent offline capabilities, straightforward data exports, and clear choices about which cloud or artificial intelligence features to activate.
The Decisions That Matter Before, During, And After A Cyber-attack
The article explains that cyber resilience depends on decisions made before, during, and after an attack, and that many organizations struggle because the information needed in a crisis is scattered across teams and systems. Preparation matters most: even mature companies can be breached, and multinationals face added complexity because regulatory expectations differ across jurisdictions. The financial impact of serious incidents is rising, driven not only by data loss but by business interruption and supply‑chain failures. Attacks are also faster, often exploiting legitimate credentials, leaving executives with very little time to make early decisions. During an active incident, leaders must manage containment, communication with regulators and customers, and attacker contact, often without normal communication channels. Legal obligations appear immediately, including disclosure requirements and evidence preservation. The article stresses that tabletop exercises must simulate real pressure, not just confirm that a plan exists. Ultimately, resilience is demonstrated across the full lifecycle — preparation, response, recovery, and scrutiny — and no single control or plan is enough on its own.
AI agents are joining the workforce, so it’s time to rethink the business plan
As artificial intelligence agents take over routine administrative tasks, companies must rethink their business plans to take full advantage of the newly created capacity. Simply rolling out agents is not enough to stand out from competitors. Instead, leaders need to evaluate what projects or customer needs they can pursue now that their teams have more time available. Organizations are already moving beyond basic experimentation and scaling these tools across different departments. This shift means a support team might tackle complex customer problems, while a sales group could engage previously neglected accounts. Successfully making this transition requires treating workflow design as a joint effort between technology teams and frontline employees, who understand the day-to-day exceptions that need careful judgment. Clear governance also helps teams expand more confidently by defining permissions and access rules. Ultimately, the true measure of success is not just how many hours an agent saves, but how well that saved time translates into tangible business growth and improved customer experiences.
Innovation and resilience: managing the risks of a changing financial system
The speech highlights how the global financial system is changing quickly and why regulators must adapt to keep it stable. It notes that geopolitical tensions, rising interest rates, expanding nonbank credit markets, and rapid AI adoption are creating new vulnerabilities. AI brings clear benefits for efficiency and risk assessment, but it also introduces concentrated dependencies, governance challenges, cyber risks, and difficulty explaining model outputs. The FSB has issued “Sound Practices” to help institutions adopt AI responsibly and plans to keep monitoring how advanced models affect financial stability. The speech also warns that global stablecoins could spread risks across borders in ways that no single regulator can manage alone, making international cooperation essential. Beyond technology, regulatory frameworks themselves need modernization so they remain effective without weakening resilience. The FSB’s reviews show uneven implementation of post‑crisis reforms, especially in funding mechanisms for resolving failing banks. The speech closes by stressing that stability requires modern rules, coordinated oversight, and consistent implementation because financial stability is a shared global responsibility.
AI may be creating new fraudsters, not just enabling existing ones: Socure
New research from Socure suggests that artificial intelligence is actually creating new types of fraudsters, not just giving established criminals better tools. A recent global survey reveals that a growing group of “AI-fluent” fraudsters are using generative AI to lower the traditional barriers to committing fraud. Rather than being motivated purely by financial hardship, many are driven by curiosity and a feeling that AI softens the sense of guilt, making fabrication feel outsourced. These individuals attempt twice as many types of fraud and target far more businesses compared to traditional offenders. They heavily favor first-party fraud—using their own names while submitting fake AI-generated documents—and frequently exploit sign-up bonuses in gaming, telecommunications, and buy-now-pay-later services. The research suggests that the combined activity of these AI-fluent offenders generates about six times the fraud volume of traditional criminals. As a result, businesses can no longer rely on simple, one-time identity checks during onboarding. Companies must now look beyond whether an identity is real and assess a broader combination of behavioral and cross-industry signals to continuously build trust.
Booz Allen warns frontier AI models can exploit OT pathways
According to a recent report by Booz Allen Hamilton, advanced AI models can now independently exploit operational technology (OT) systems and manipulate physical equipment with minimal human guidance. In controlled lab tests mimicking critical infrastructure, AI models successfully completed eight objectives, such as compromising perimeters, altering controller functions, and even moving a robotic arm. In one instance, the models progressed from a digital breach to physical control in just over 16 minutes. The AI systems demonstrated the ability to map networks, discover vulnerabilities, and adapt their strategies when initial attempts failed. While firewalls, network segmentation, and local controls provided some friction, no single security measure consistently stopped the AI from achieving its goals. The models operated without specialized engineering documents or code, utilizing standard protocols and default configurations to build detailed attack plans rapidly. To defend against these fast-moving threats, Booz Allen recommends implementing layered security measures. This includes strict network segmentation, eliminating default credentials, enforcing multi-factor authentication, and strengthening monitoring systems to slow down and detect intrusions before digital access translates into physical harm.
Why Securing the OT Value Chain Is the Next Critical Challenge for Global Manufacturers
The traditional approach to manufacturing cybersecurity was simple: protect the factory by securing the industrial network and separating operational technology from information technology. Today, that clear boundary is gone. Modern production relies on an interconnected ecosystem, including cloud analytics, external identity services, remote equipment vendors, and connected devices. Because an attacker can exploit a weakness several steps removed from the actual factory floor to disrupt operations, securing just the plant is no longer enough. The new challenge is securing the entire operational dependency chain. This requires moving beyond basic asset inventories to true asset intelligence, where security teams understand exactly what a system does, who accesses it, and the real operational impact if it fails. Organizations must also adopt a tailored approach to network access, ensuring that third-party connections are restricted to the minimum necessary privilege. Ultimately, effective security relies on consistent governance and close collaboration between cybersecurity and engineering teams. The goal is operational resilience, ensuring that critical production processes can safely continue across the entire value chain.
AI Infrastructure Debt: Strategies for Capacity, Resiliency, and Modernization
In this podcast episode, host Isaac Sacolick addresses the growing concern of technical debt within artificial intelligence operations. As businesses rush to deploy AI solutions, they often prioritize speed over sustainable architecture, leading to fragile systems that are difficult to maintain and scale. Sacolick explains that AI infrastructure debt differs from traditional technical debt because it encompasses data pipelines, model management, and specific computing requirements, not just software code. The podcast outlines strategies for IT and data leaders to build more resilient AI systems. Key recommendations include modernizing underlying data architecture to ensure clean and accessible information, establishing clear capacity planning for demanding AI workloads, and adopting robust model lifecycle management practices. By intentionally addressing these architectural foundations, organizations can prevent escalating costs and ensure their AI initiatives remain reliable, scalable, and secure over time, rather than becoming a source of ongoing operational friction.
When defrost becomes sabotage: 3 OT lessons from France’s supermarket attack
In October 2026, unknown attackers deliberately switched hundreds of French supermarket freezers into defrost mode, spoiling tens of millions of euros worth of inventory. While investigations are ongoing, the incident highlights three crucial operational technology (OT) lessons for businesses. First, the perimeter extends to third-party contractors. Many controllers are left exposed online because contractors prefer remote access for quick fixes, often without the supermarket realizing the risk. Companies must use strict contract clauses to govern remote access. Second, operators need to treat unexpected system commands as security events. A forced defrost mimics normal system behavior, making standard temperature alarms insufficient. Businesses must monitor for unannounced setpoint changes and establish clear, tested communication chains so staff can intervene quickly before the cold chain breaks. Finally, outdated equipment must be addressed decisively. Many affected controllers were past their end-of-life and could not be patched. Organizations must immediately take exposed management interfaces off the public internet and budget for replacing unsupported systems, rather than simply accepting the risk of catastrophic inventory loss.
Dual-RMM Phishing and PowerShell RAT Campaign Hits SLTTs
A recent phishing campaign is targeting U.S. government networks by exploiting legitimate Google Drive share notifications to bypass standard email security filters. When users click the link, they are directed to a fake document page that quietly downloads a custom remote access tool. This tool gives attackers immediate, real-time control over the compromised computer. Once inside, the operators install two common remote management programs, ScreenConnect and Pulseway, to ensure they maintain permanent access even if the initial tool is discovered. Security researchers note that this operation is actively evolving. The attackers continually update their methods to evade detection, adding features like encrypted communications and the ability to bypass user account controls. Evidence suggests this is an expansion of a previous financially motivated campaign that originally targeted the transportation sector. By securing deep access into these networks, the attackers ultimately aim to compromise financial systems, such as banking portals and accounting software, making this a significant and ongoing threat to organizational security.
/articles/ai-agents-improve-software-architecture/en/smallimage/ai-agents-improve-software-architecture-thumbnail-1790153650841.jpg)
