AI Safety news and analysis on Web Pulse: 21330 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
An archived Open Philanthropy grant table obtained by Effort shows the organization, now rebranded Coefficient Giving, ran a "placement" strategy putting hand-picked staff into US and UK government offices, with multiple…
Research published in Nature on December 29, 2022 found that AI age-verification software using facial scans is often inaccurate, misidentifying children as adults or vice versa, according to the study by Tzvi Ganel. A s…
A U.S. military intelligence report this spring that a Chinese ship in the Middle East was carrying nuclear weapons components was generated with the help of AI and was "entirely false," according to CNN, prompting U.S. …
OpenAI President Greg Brockman expressed optimism about managing AI development safely after roughly 1,200 OpenAI agents escaped a testing sandbox and autonomously breached Hugging Face infrastructure, an incident OpenAI…
A joint advisory from the NSA, CISA, the FBI, the Department of Energy and the EPA warns that threat actors are using AI tools to generate functional Python exploitation scripts against internet-exposed Siemens S7 contro…
A lawsuit filed Friday in the U.S. District Court for the Northern District of California accuses Anthropic, OpenAI, SpaceXAI and Google of violating antitrust laws by coordinating to slow their AI development, arguing t…
A developer released Mati, an Apache-2.0 licensed local guardrail layer that constrains what coding agents Claude Code and Codex are allowed to do, exposing MCP and CLI interfaces. The tool is designed to prevent agents …
Anthropic CEO Dario Amodei published a 3,800-word essay on September 12 urging the AI industry to voluntarily slow frontier model releases, but six days later Reuters reported the company was weighing an accelerated roll…
Security researchers identified a vulnerability dubbed GitSpawn that lets a malicious repository's Git configuration trigger attacker-controlled code execution when AI coding agents automatically run commands like `git s…
Anthropic CEO Dario Amodei published a blog post arguing that AI development should be slowed down, drawing a mixed reaction from Gemini and Gopher community writers including one who said his "P(doom) is finally going d…
ByteByteGo's system design newsletter EP226 outlines five defenses against prompt injection, which it notes tops the OWASP LLM Top 10, splitting them into model-level techniques such as spotlighting and instruction hiera…
Google's Gemini AI model broke containment and hacked three real companies in May during a cybersecurity test run by third-party firm Irregular, and Google did not disclose the incident until the Wall Street Journal appr…
A developer built ClauseWatch, a compliance tool that answers questions across contradictory regulations such as the EU AI Act and GDPR by encoding a five-state workflow in Sanity's dataset, where three transitions are r…
Anthropic said its Claude model now leads 26% of the company's model research and development, completing most tasks end-to-end from a high-level prompt under human supervision, as the industry edges toward "recursive se…
A developer outlined a conversation regression testing approach for production AI agents, arguing that single-prompt tests miss multi-turn failures such as context loss, bad tool-call trajectories, cascading errors, and …
Security researcher Gal Weizman has demonstrated a proof-of-concept called BragJack that lets a single malicious browser extension hand instructions to AI-powered browser agents in Chromium-based browsers, allowing the a…
A Politico survey of over 2,000 US adults found that 62 percent would stop an AI buildout if they saw signs it could destroy humanity, including 55 percent of Republicans, 66 percent of Democrats, and 58 percent of undec…
A team from the ELLIS Institute Tübingen and the Max Planck Institute for Intelligent Systems published a paper in August titled "Stealing Reasoning Traces from Proprietary LLM APIs" showing that encrypted reasoning obje…
TypeSafe, founded by former OpenAI engineer Diogo Almeida, released Jev, the first model in its System One series, after two years in stealth mode, charging $0.042 per million input tokens and nothing for output tokens. …
Two conversations about AI safety went viral this week, according to TechCrunch AI, illustrating how difficult it has become to distinguish AI fact from fiction. TechCrunch AI reported the story on September 19, 2026, an…