AI Safety news and analysis on Web Pulse: 12542 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Arcadia Alignment's research reveals that current AI model organisms used to study alignment pathologies suffer from degraded coherence, instruction-following, and reasoning, making them poor proxies for real deployment …
U.S. lawmakers are struggling to keep pace with rapid AI advancements, introducing over 1,300 state-level AI bills this year while lacking independent technical expertise. Programs like the Conservative AI Policy Fellows…
Meta's Applied AI unit, a 6,500-person team created to power CEO Mark Zuckerberg's AI ambitions, is in turmoil after elite engineers were forcibly reassigned to data-labeling tasks, with employees describing the environm…
Dean Ball will join OpenAI on July 6 as Head of Strategic Futures, a new team focused on frontier AI policy including catastrophic risk, recursive self-improvement, labor market impact, and government relations. Ball wil…
A KFF poll released Wednesday found that 29% of U.S. adults now use AI tools like ChatGPT, Google Gemini, and Claude for health advice monthly, nearly double the share from June 2024. The survey also showed 30% use socia…
A survey of 4,000 American consumers found that 39% store old electronics instead of recycling or reselling, driven by data security fears and lack of knowledge. Researchers from the National Science Foundation-funded st…
The Senate Judiciary Committee unanimously cleared the No Fakes Act by voice vote on Thursday, creating a legal right for individuals to authorize use of their voice and likeness in AI-generated content and demand takedo…
A developer documented a secret scanning setup that prevents credentials from entering the context window of Claude Code, an AI coding assistant. The system uses trufflehog with 700+ detectors to intercept secrets at fou…
Cory Doctorow argues that the concept of AI digital sovereignty risk is a fallacy, using the analogy of blockchain's failure to solve trust problems. He asserts that adding AI to a problem does not eliminate the underlyi…
A developer built Proofline, a 5-skill Markdown pack that catches fake-ready output from AI agents before it becomes a release or handoff. The tool works as a review route after agent output, asking questions about missi…
RippleX has implemented a security-first approach for the XRPL Lending Protocol and Single Asset Vault (SAV) amendments, subjecting them to ten phases of review including AI-assisted security reviews, multiple independen…
Democratic operatives Shaunna Thomas and Leah Hunt-Hendrix launched the Guardrails Alliance, a super PAC backed by tech workers and labor unions, to support AI legislation with $5 million. The PAC aims to counter the inf…
Sebastian Kurz and Shalev Hulio's AI cybersecurity startup DREAM raised $260 million in Series C funding at a $3 billion valuation, led by Bicycle Capital and Group 11. The company sells sovereign AI cyber defense system…
Accenture announced Thursday it will acquire a majority stake in industrial cybersecurity firm Dragos for $3.25 billion and purchase two smaller security companies, runZero and NetRise, for a total of $4.18 billion. The …
Anthropic CEO Dario Amodei manages only one direct report—his chief of staff—while all other senior leaders report to his sister and company president Daniela Amodei, a structure he calls 'incredibly freeing' that allows…
Google's AI Overviews feature is presenting fictional entities from the SCP Foundation horror fiction universe as real, without acknowledging their fictional nature. Searches for SCP-565 and SCP-426 return AI-generated s…
Researchers tested 400 ultra-cheap AI API proxy services offering GPT and Claude access at 90% off, finding that one drained crypto from a wallet while others injected malicious code or stole cloud credentials. The repor…
The TAKE IT DOWN Act, targeting non-consensual deepfake imagery, was signed into law by President Trump on May 19, 2025, after passing the Senate unanimously and the House with a 409-2 vote. The bill requires platforms t…
NILScript introduces an OpenAPI-like standard for agent actions that prevents unauthorized writes by enforcing a propose-approve-commit-rollback workflow. In 4,216 prompt-injection evaluations across two models, the syst…
A developer built a scanner that fires prompt-injection probes at a self-hosted AI agent and tested it across five model backends, finding leak rates ranging from 0% to 90% depending solely on the model. The scanner dete…