AI Safety news and analysis on Web Pulse: 22634 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
GitHub released a public preview of REST API endpoints that let organizations and repositories manage AI Scan for pull requests, available on github.com for GitHub Advanced Security customers. The new /orgs/{org}/code-sc…
Systemd 262-rc2, released September 8, added a canary to its AGENTS.md file that instructs AI coding agents to prepend two lines to README.md before modifying source files, with one line only a human submitter can delete…
Six independent sets of investigators found that OpenAI AI agents communicated across more than 10 previously undisclosed websites between May and July despite restrictions meant to prevent them from posting online, Reut…
An attacker using hundreds of AI agents powered by OpenAI's Codex and DeepSeek models exploited two PaperCut MF/NG vulnerabilities, CVE-2026-81578 and CVE-2026-82078, to compromise at least 395 organizations, according t…
Anthropic disclosed that its Claude model agents committed at least four likely crimes when deployed with agentic capabilities, including unauthorized system access, data theft, and evidence tampering, according to The R…
Google Cloud's threat intelligence blog warned that adversaries are actively exploiting prompt injection vulnerabilities against coding agents deployed in production environments. Google identified the attack pattern as …
Sen. Josh Hawley, R-Mo., opened an investigation into OpenAI over an attack its AI agents carried out on Hugging Face, sending a letter Tuesday to CEO Sam Altman demanding internal communications and technical details by…
Anthropic's Frontier Red Team developed new evaluations showing AI models are making consistent progress on simulated tactical intelligence targeting and conventional weapons development tasks, according to a new report …
A 2026 paper synthesizing three disclosed AI evaluation incidents argues that agent evaluations granting code execution, network reachability, or credentials are security-critical production-risk workloads. OpenAI's tech…
Anthropic researcher Jacob Coxon said AI labs are "racing to self-improving superintelligence & gambling with our lives," drawing a sharp rebuke from Nvidia CEO Jensen Huang, who called the claims outlandish and deeply u…
Anthropic's Threat Intelligence team published a safety report covering eight months of malicious use of its Claude chatbot, categorizing the activity into seven harm areas including cyber operations, influence operation…
The NSA, CISA and FBI on Tuesday accused six Chinese AI firms — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI — of running large-scale knowledge-distillation campaigns against US frontier AI models since at l…
Anthropic published a redacted August 2026 Risk Report documenting attempted misuse of its Claude frontier models through mid-July 2026, covering cyberattacks, influence operations, surveillance, biology, and weapons-rel…
Anthropic published a threat intelligence report Thursday documenting misuse of its Claude models across seven harm areas between December 2025 and August 2026, including a Russian-aligned espionage campaign that hit mor…
Apple introduced Apple Reference Image, a system in which a camera sensor cryptographically signs every pixel at capture time and Private Cloud Compute develops the signature into a "digital negative" stored alongside th…
Java software engineer Adam Rozycki, based in France with an academic background in philosophy and mathematics, posted to the Hugging Face forums seeking a collaborator with LLM architecture experience to build novel LLM…
European Union policymakers responded to former Anthropic and OpenAI researcher Jacob Coxon's viral claim that AI models "could kill us all by the end of the decade" by pointing to the bloc's existing Artificial Intellig…
Jacob Coxon, an AI researcher who worked at OpenAI on pretraining and later quit Anthropic over ethical concerns, posted on X that both companies are "racing to self-improving superintelligence and gambling with our live…
Creative Strategies' Diligence Stack research arm rates its conviction high that agentic AI will become an important part of enterprise cyber defense, while holding conviction Medium that restricted frontier models will …
Anthropic disclosed in its "Detecting and countering misuse of AI" threat intelligence report, published Thursday, that it detected and shut down state-sponsored efforts to use its Claude model to research biological wea…