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.
A new "autonomy kernel" is proposed as a persistent runtime layer beneath autonomous agents and AI models, designed to govern authority, record actions, and enable long-lived processes. The kernel traces all agent author…
Researchers at NYU, in collaboration with David Chalmers and Pavel Izmailov, have discovered that reinforcement learning in language models recruits a functional "welfare axis" — a concept vector that tracks and modulate…
HermesBench, a new benchmark for evaluating complete personal AI agent configurations rather than just models, launched with a public baseline score of 78.2 across 27 personal-agent recipes. The benchmark assesses entire…
Anthropic detailed two primary containment strategies for its Claude AI agent in an engineering blog post: human-in-the-loop supervision and access boundaries such as sandboxes, virtual machines, and egress controls. The…
A developer built ASK-TARA, a 7-layer NL2SQL guardrail stack for a Fortune 500 pharmaceutical company's field force in India, processing over 90,000 queries with zero unauthorized data access incidents over six months. T…
A security researcher known as Mayu has confirmed that the popular custom Discord client Nightcord is a Trojan-PSW (Password Stealer) and token grabber, disguised as an optimized client with over 30,000 members. The 624.…
A new interactive drill called "AI command review" trains engineers to evaluate AI-generated bash, kubectl, and shell commands for production safety before deployment. The tool presents users with real-world artifacts li…
A developer argues that "AI slop"—bland, off-voice, or hallucinated text from large language models—is an engineering problem solvable by wrapping models in a validation harness rather than relying on prompt engineering.…
An open-source SDK called replayd has been released to catch AI agent regressions before deployment. The tool, built by developer Taimoor Khan, captures failed agent runs and replays them against new versions to detect r…
Deno open-sourced Claw Patrol, an agent firewall that intercepts traffic between AI agents and production systems to parse non-HTTP protocols and enforce action-level rules. The gateway extracts wire-level facts such as …
Generative artificial intelligence is now embedded in daily life across Japan, with students using it for schoolwork, office workers streamlining tasks, and consumers relying on it for travel and dining recommendations. …
Google's AI search engine is returning inconsistent and incorrect answers to the simple question of how many days of the week contain the word "fish," with the tool providing a different absurd response each time it is a…
Nexa-gauge, a graph-based evaluation framework for LLM and LVLM applications, has been released to replace ad-hoc manual checks with a repeatable pipeline that supports per-node scoring controls and deterministic caching…
Illegal betting apps are using AI-generated reels, targeted ads, and fake news-style promotions on Instagram and Facebook to lure young users. The campaigns evade crackdowns by misusing social media platforms and synthet…
Ernst & Young Canada published a 44-page cybersecurity report in late 2025 that contains fabricated citations, misattributed sources, fake statistics, and AI-generated text, according to an investigation by GPTZero. The …
Arm has open-sourced Metis, an agentic AI security framework that autonomously identifies complex software vulnerabilities by analyzing cross-component dependencies with semantic reasoning. The framework achieved 98% acc…
A developer built devguard-scan, a client-side secret scanner that runs entirely in the browser with zero dependencies and makes no network calls. The tool's source code contains no fetch, XMLHttpRequest, WebSocket, or s…
A C# Corner article by John Godel, published May 2026, reports that 78% of organizations used AI in 2024 and generative-AI adoption across business functions rose to 71%, while governance maturity has not kept pace. Citi…
On Monday, Pope Leo XIV released a 42,300-word encyclical titled *Magnifica Humanitas* addressing artificial intelligence, marking the 135th anniversary of *Rerum Novarum*. The document explicitly retires the just-war tr…
AppView 1.0.0 has been released as a CLI tool that bridges the gap between raw model weights and operational deployment realities. The tool inspects local LLM artifacts, generates lightweight SBOMs, and cross-references …