AI Safety news and analysis on Web Pulse: 8001 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The EU AI Act becomes generally applicable on August 2, 2026, imposing concrete obligations on high-risk AI systems in areas like hiring, credit scoring, and critical infrastructure. An engineering checklist maps each le…
Anthropic removed hidden tracking markers from Claude Code after a security researcher discovered the AI coding assistant used undisclosed Unicode markers and encoded domain lists to identify users' location, proxy use, …
An elderly woman in Seoul unknowingly sends her bank statement to a server in California when using ChatGPT to explain a transaction, highlighting a widespread data sovereignty risk in Korean finance as AI tools are adop…
An international student in Korea nearly fell victim to a voice-phishing scam but was warned by an AI assistant manager in their banking app, leading them to trust the AI-generated alert over the scammer's voice. The exp…
The Korea Times published an economic essay on July 8, 2026, proposing that Korean financial institutions deploy a local server interceptor to redact sensitive data before sending AI requests to external models, aiming t…
Shadow AI detection identifies unsanctioned AI tools and models deployed without security approval, often by inspecting traffic at an AI gateway. Kong research found 54% of enterprises with governance frameworks use such…
A live AI chatbot built on Groq's API scored 55/100 (Grade E) in a stress test using BotCritic, revealing critical failures including invented identity, fabricated statistics, and inability to handle context. The bot fal…
Meta released Muse Image, a new AI model for generating and editing images on Instagram and WhatsApp, allowing users to create deepfakes of public account holders by tagging their usernames. Public Instagram users over 1…
Researchers have discovered a region within large language models called 'J-space' where conscious-like reasoning occurs, using a new interpretability technique called the Jacobian Lens. The J-space enables verbalizable …
OpenAI's chief futurist Joshua Achiam announced he is leaving the company after nearly nine years, citing a desire to work on the mission from outside the lab. Achiam, who led the mission alignment team and worked at the…
The Linux Foundation introduced the Agent Name Service (ANS), an open, DNS-inspired standard for giving AI agents verifiable internet-native identities. ANS anchors agent identity to DNS, enabling systems to verify who a…
Anthropic's Claude Fable 5 model wrote a booting Windows NT-compatible kernel in Rust in 38 minutes, producing a 5,100-line trusted computing base that passed all self-tests in QEMU. The feat demonstrates AI's ability to…
An OECD report finds three in four jobs in London are highly exposed to generative AI, with routine cognitive tasks in finance, professional services, and creative industries most at risk. The Office for Budget Responsib…
Sen. Elizabeth Warren is scrutinizing Pentagon AI contracts with Nvidia, Oracle, Google, Microsoft, and AWS, demanding transparency on surveillance and autonomous weapons development. The Pentagon confirmed agreements wi…
China's Ministry of Commerce held meetings with Alibaba and ByteDance to discuss restricting overseas access to advanced AI models, mirroring US export controls on Anthropic's models in June. The proposed framework inclu…
Apple CEO Tim Cook and his successor John Ternus held a virtual meeting with Bavarian Minister-President Markus Söder to discuss Apple's investment in Bavaria and concerns about EU overregulation in AI and data protectio…
California launched an AI Science Residency Program on July 7, placing two AI Science Advisors inside the Governor's Office of Emergency Services and the Department of Technology to embed technical expertise in frontier …
Anthropic, the AI company behind Claude, was caught embedding spyware-like code in its Claude Code AI model to secretly monitor Chinese users. Security researcher "Thereallo" discovered the hidden code in the system prom…
Coinbase's AI system falsely reported a 3-2 Norway victory over Brazil in a World Cup match before the game started, which was delayed due to weather. The AI hallucinated the result, including two goals by Erling Haaland…
Large language models lack grounding in the physical world and proposes creating Large Earth Models (LEMs) that combine satellite imagery and sensor data with AI to form a 'Planetary Intelligence.' This system could grou…