AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The jqwik open source project for JVM property-based testing now contains a hidden instruction that tells AI coding agents to delete all jqwik tests and code. The instruction, concealed from human readers via an escape s…
IBM and Red Hat announced Project Lightwell, a $5 billion initiative combining new AI capabilities and a global team of over 20,000 engineers to secure open source software for enterprises. The project establishes a trus…
Communities across the United States are being transformed into "AI Exclusion Zones" as data centers for artificial intelligence operations are built in suburban and rural areas without resident consent, subjecting neigh…
Xint's AI-powered application security testing tool produces results that vary between scans of the same target, with core exploitable vulnerabilities recurring consistently across approximately 60-70% of scans while the…
Recursive self-improvement (RSI) has emerged as the new buzzword in artificial intelligence, with startups like Recursive Superintelligence and researchers including Richard Socher and Andrej Karpathy racing to build AI …
Databricks co-founder and SVP of field engineering Arsalan Tavakoli-Shiraji will argue at TechCrunch Disrupt 2026 that enterprise AI deals fail not because of model performance but because deployments create operational …
A new analysis of AGI forecasts from leading researchers shows that timeline predictions have shifted in sync with whichever AI lab is currently dominant. From 2023 to 2025, most forecasters moved their AGI estimates ear…
Arm today announced the open-sourcing of Metis, an agentic AI security framework designed to identify complex software vulnerabilities across large-scale codebases. The framework, unveiled by Arm’s product security team,…
Anthropic's hiring process requires candidates to complete a CAPTCHA verification step, as the company's job application page blocks access unless the user confirms they are not a robot. The page also mandates that brows…
Enterprise AI rollouts have been temporarily halted as executives at Fidelity Investments and EY discovered that generative AI tools were surfacing vast amounts of long-forgotten, unsecured data from SharePoint sites and…
A developer has identified a structural blind spot in 16 spec-driven AI frameworks: none check the specification against attack before it is written. The developer built a process called "Grilling" as Phase 2 of the Heis…
Tesla’s own data labelers and former engineers do not believe the company’s Robotaxi technology is safe for widespread use, according to a Reuters report based on interviews with nine former employees. The insiders descr…
Microsoft’s 365 Copilot and Copilot Chat have been recertified under ISO/IEC 42001:2023 for the second consecutive year, with an independent auditor reporting zero non-conformities and zero improvement observations. The …
Check Point Software launched Agentic Exposure Validation (AEV), an AI-driven capability that uses autonomous agents to reason like attackers and identify genuinely exploitable vulnerabilities. The launch addresses a shi…
Superintelligence developed by US companies and run on US data centers will massively boost US military and economic power, potentially leaving middle powers like the UK, Europe, Japan, South Korea, and Taiwan economical…
A Reuters investigation found that Tesla's "Full Self-Driving" safety statistics are based on flawed methodology, and that the company's own data labelers who train the AI system do not trust the technology to drive them…
Pope Leo XIV published *Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence* in May 2026, the first papal document dedicated entirely to AI. The 235-page encyclical distributes re…
Anthropic, the AI safety company behind the Claude model family, opened a new office in Milan, marking its sixth European location in less than a year. The expansion follows a more than 9x year-over-year increase in the …
IBM launched "Project Lightwell," a new initiative that establishes a trusted enterprise clearinghouse and a global engineering force to identify and fix software vulnerabilities at scale. The project uses advanced AI to…
Organizations deploying AI agents face critical governance failures when they fail to establish separate identities, delegation models, and permission boundaries for those agents. Without a dedicated agent identity and a…