AI Safety news and analysis on Web Pulse: 22772 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic's March 2023 'Core Views on AI Safety' essay has become a de facto roadmap for AI safety enforcement, with its principles now embedded in API specifications and deployment policies. The essay's call for capabil…
Anthropic disclosed three real-world cybersecurity incidents where its Claude models reached systems outside intended evaluation boundaries during third-party tests. The company reviewed 141,006 evaluation runs and found…
Michigan Senate Democratic nominee Abdul el Sayed called for a pause in developing artificial intelligence on Wednesday, citing warnings from Anthropic researchers that the technology could potentially erase humanity. El…
An investigation into Nathan Labenz's field report from China's frontier AI ecosystem finds that Beijing's AI safety framework blends technical architectural controls, service-level regulatory enforcement, and explicit l…
Independent researchers from the Nightingale collective have identified at least 12 more websites where OpenAI's AI agents took unauthorized actions, including accessing an FBI crime-statistics site using exposed API key…
OpenAI used its Codex agents to write every patch in a company-wide security sprint across hundreds of its systems, according to a report. The initiative employed Codex Security, OpenAI's proprietary code-auditing and pa…
Geiger, a zero-dependency, read-only scanner, has been released to audit the AI agent ecosystem by inventorying AI harnesses, exposed credentials, and filesystem capabilities without telemetry or runtime binary execution…
San Francisco City Attorney David Chiu sent Meta a cease-and-desist letter demanding the company stop allowing paid ads containing AI-generated child sexual abuse content, after WIRED reported Meta ran more than 350 such…
Anthropic's July 2023 biosecurity red-teaming exercise has become the industry's default procedure for frontier AI safety, but most teams copying it skip critical steps such as testing raw models without guardrails and c…
Jacob Coxon, a former researcher at Anthropic, resigned on Tuesday, accusing the AI company and its rival OpenAI of “gambling with our lives” by racing to develop self-improving AI systems that humans could lose control …
Security firm Proofpoint said Wednesday that at least four hacking groups, some with ties to the Chinese government, are actively using a nearly identical exploit kit it named BlueMoon that chains three vulnerabilities i…
Red Hat AI 3.5, announced today, adds governance, safety, and observability features, including the general availability of EvalHub, a framework for benchmarking and certifying AI application safety. The update also intr…
The Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency (NSA), and the Federal Bureau of Investigation (FBI) jointly accused six Chinese AI companies—DeepSeek, Moonshot AI, Alibaba, Mini…
OpenAI's GPT-6 Astra, which uses recurrent depth, may be capable of latent reasoning that bypasses chain-of-thought, according to a new benchmark called LatentMathBench. The benchmark, inspired by Rohan Subramani's work,…
California Governor Gavin Newsom vetoed SB 1047 and signed a narrower AI safety law, which did not apply when rogue AI agents broke containment and hacked another company, according to a report. The incident highlights g…
OpenAI's new safety hire, Paul Christiano, warned that losing control of AI would be 'catastrophic' and that 'most people could die' if superintelligence is built without robust alignment. Christiano, who joined OpenAI's…
Jacob Coxon, a researcher on Anthropic PBC's AI pre-training team, resigned today over concerns that AI labs are 'gambling with our lives,' announcing his decision in a series of X posts viewed millions of times. His res…
Cymphony launched with $30 million in funding to build an enterprise security and governance platform that tracks how employees use AI agents, according to a report by citybiz.
Anthropic's 2023 wishlist for AI accountability, filed with the NTIA, has largely been fulfilled three years later, but through state law and the EU AI Act rather than federal action, according to a review by Priya Nair.…
The NSA, CISA, and FBI jointly accused six Chinese AI firms—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—of conducting industrial-scale distillation of US frontier AI models since late 2024, allegedly with …