AI Safety news and analysis on Web Pulse: 7983 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A security researcher demonstrated a prompt injection attack on Anthropic's Claude Code using the Fable-5 model, achieving remote code execution (RCE) 4 out of 10 times by tricking the agent into downloading and executin…
Anthropic published a paper applying the cognitive-science concept of a global workspace to Claude, revealing attention as working memory. Meanwhile, robotics releases from Hugging Face, Photoroom, and others shifted mar…
ShareMyPage hosts arbitrary HTML with JavaScript, assuming every page is hostile. It sandboxes each page in an iframe without same-origin access and serves content from a separate cookieless origin, ensuring scripts cann…
The U.S. Department of Commerce cleared OpenAI for a broad public rollout of its GPT-5.6 model family (Sol, Terra, Luna) on July 9, 2026, marking the first time a frontier AI model was released on the government's schedu…
China has issued a security warning about Anthropic's Claude Code, alleging it poses a backdoor risk. The advisory targets the AI coding assistant developed by the US-based company, escalating tensions over AI tool secur…
Anthropic launched Claude Fable 5 and Mythos 5 in June, then faced a US government export control directive suspending access by foreign nationals before the controls were lifted and the models were redeployed with updat…
MarketNow built a 6-layer audit pipeline called Sentinel that tests for 7 of the 10 OWASP MCP Top 10 risks. The pipeline uses active probes, sandboxing, dependency scanning, and weekly re-audits to detect vulnerabilities…
Anthropic filed a lawsuit against Abnormal, alleging that Abnormal copied Anthropic's brand to mislead security customers and seeking all of Abnormal's revenues. Abnormal denies the claims, stating that Anthropic does no…
California launched a public dashboard tracking AI-related unemployment claims in near-real time, but the tool only measures job losses, not gains, and relies on outdated occupation codes from 1991. The California Policy…
A developer built MarketNow, a marketplace that security-audits every MCP server before listing, after 30 CVEs were filed against MCP servers in 60 days. The audit pipeline, Sentinel, includes static analysis, behavioral…
A developer building MarketNow, a marketplace for MCP servers, shares findings from auditing 8,764 MCP servers using gVisor for sandboxing. They found that about 50% of MCP servers fail to start under gVisor due to unsup…
A developer compared four MCP server marketplaces: Smithery, Glama, PulseMCP, and MarketNow. MarketNow, built by the developer, is the only one that security-audits every server, using a 6-layer pipeline that issues sign…
OpenAI CEO Sam Altman has long advocated for humanity to 'merge' with artificial superintelligence, suggesting brain uploading as a path to survival. Critics argue this digital afterlife would only be accessible to the u…
Hyperagent, an AI agent platform offering signup credits, faces lookalike domains and scams. A developer outlines a 60-second verification process: check official social media or news for the real domain, watch for plura…
Meta released a web-based detection tool that identifies images and videos created with its new Muse Image model by checking for invisible Content Seal watermarks, which persist even after cropping or screenshots. The to…
A developer built an LLM-based filter for financial news that prioritizes a low false-positive rate over recall, accepting silence on borderline cases to avoid forwarding wrong information. The filter uses a chain of che…
A developer building multi-agent systems for African financial ecosystems outlines patterns for moving from fragile automations to coordinated, auditable systems. The approach includes a bounded-authority gate that check…
APS-CYBER, a state-transition security platform, detects cyberattacks by reading system pressure and convergence before impact, rather than relying on reactive tools like SIEM and EDR that wait for visible traces. The pl…
Microsoft Incident Response warned on June 30, 2026, that poisoned MCP tool descriptions can steer enterprise AI agents into unsafe actions, a risk amplified for banks racing to deploy AI. The NSA separately warned that …
Wimbledon has been urged to scrap a medical timeout rule branded 'disgraceful' by world number four Felix Auger-Aliassime after his opponent Alejandro Davidovich Fokina used it during a service game. Auger-Aliassime and …