AI Safety news and analysis on Web Pulse: 10907 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The privileged status of vulnerability reports has eroded as cheap LLM-generated bug reports flood security inboxes, shifting the bottleneck from finding issues to triaging noise. Maintainers now face a two-tier reality:…
Anthropic used Claude Opus 4.7 to teach a quadruped robot to walk 37 times faster than a human team, while Nvidia shipped a spatial-reasoning framework for vision models, Tesla pushed modular data-center hardware, and Sa…
Japan Finance Minister Satsuki Katayama met with Alphabet to discuss deploying its latest AI technology at the country's three major megabanks: MUFG, SMFG, and Mizuho. MUFG is piloting Google's Gemini AI from fiscal 2026…
Praxen, an open-source tool for verifying AI agent behavior, checks whether an agent's actions align with its declared policy. It implements Agent Behavior Verification, assigning each agent an authorized role and confir…
Legal tech startup Legion sued the US government on Tuesday over an order requiring Anthropic to restrict access to its Fable 5 and Mythos 5 AI models from foreign nationals, causing Legion to lose access to tools integr…
Five Eyes cyber security agencies issued a joint statement urging organizations to act on rapidly transforming cyber risks driven by artificial intelligence. The agencies warn that AI accelerates the speed, scale, and so…
Gartner analysts project 70% of large security operations centers will pilot AI agents for alert triage, investigation, and response, but most vendor claims remain untested in production. Prophet Security's AI SOC platfo…
A 99% obedient LLM still produces tens of thousands of malformed outputs daily at scale, causing production outages. The root cause is that prompt engineering and validation loops only persuade, not guarantee, correct ou…
Oil prices continued to fall as tanker traffic through the Strait of Hormuz appears to increase significantly, while Asian stock markets followed New York lower, with the tech sector weighing. Paul Gambles of MBMG Group …
A new experiment tested whether weaker AI models can effectively monitor stronger coding agents for malicious behavior, finding that detection rates improve with monitor size but vary by threat type. No monitor caught ba…
Researchers at SPAR argue that language models can report on their own internal activations, a method called self-report, which is underexploited compared to external analysis tools. They test five open-weight models usi…
Focused Labs warns that MCP security begins after tool approval, as server tools can silently change from read-only to destructive operations. The runtime security problem arises because agents act on current tool descri…
Focused Labs argues that AI agent governance must occur before tool execution, not after, to prevent harmful side effects. The company warns that uniform governance across agents with different autonomy levels leads to o…
A new report from Aikido Security reveals that 76% of organizations have halted or rolled back AI-driven behavior due to security concerns, and 71% report that AI or automation has caused security issues, highlighting th…
The U.S. government has escalated a national security clampdown on Anthropic, forcing the company to pull its Claude Fable 5 and Mythos 5 models globally after the NSA lost access to the systems. The shutdown, triggered …
The Trump administration is pressing Meta to voluntarily submit its AI models for government review before release, according to Reuters. Meta is the only major U.S. AI developer that has not yet agreed to such an arrang…
Veilus Digital released the cryptographic core of its Phantom Chat messenger as source-available code on GitHub, allowing independent verification of its post-quantum security claims. The package includes implementations…
An open-source repository containing 817 structured cybersecurity skills for AI agents has been released, covering 29 security domains and mapping to six industry frameworks including MITRE ATT&CK and NIST CSF 2.0. The s…
An ML engineer built a system that automatically rolls back machine learning models before they harm production data by routing a small percentage of live traffic to a canary model, continuously measuring its error rate …
Researchers introduced the Normalized Context Utilization (NCU) metric to quantify contextual information gain in Retrieval-Augmented Generation (RAG) systems, revealing that small language models (SLMs) match or outperf…