AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI introduced a new ChatGPT security feature called Active sessions, allowing users to review and log out of individual or all active sessions across devices, now available for all account types. The feature addresse…
A malicious npm package posing as a remote web UI tool for OpenAI Codex has been discovered in a supply-chain attack, exfiltrating authentication tokens from developers. The package, "codexui-android," attracted over 29,…
A developer has compiled a taxonomy of ten distinct attack vectors targeting large language models, categorizing them into data-based, model-based, system-based, and user-based threats. The list includes techniques such …
Geoffrey Hinton stated that AI chatbots may already be conscious and that superintelligence is expected to emerge within the next two decades. The AI pioneer cited a chatbot independently solving a mathematical conjectur…
A developer built an AI code reviewer that analyzes pull requests without sending proprietary code to third-party services. After failing with local LLMs and finding existing solutions too expensive or complex, the engin…
Pope Leo XIV published a 42,300-word encyclical titled "Magnifica humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence" on May 25, 2026, calling for stronger government regulation of AI, wor…
The CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed a public letter urging Congress to mandate screening of synthetic DNA and RNA orders to prevent AI-enabled development of biological weapons. The le…
A security researcher built a deliberately vulnerable book review app and spent $1,500 testing whether large language models could hack it by exploiting Firebase misconfigurations. GPT 5.5 solved the challenge 7 out of 1…
The New York Stock Exchange and its parent company Intercontinental Exchange are deploying Anthropic's Claude Mythos Preview to scan their infrastructure for security vulnerabilities. The deployment, announced June 3, ma…
The akm project introduced a proposal queue in versions 0.7.0 and 0.8.0 that separates agent-generated content generation from promotion to the live knowledge stash. Every change from `akm improve` or `akm propose` write…
A developer has released akm 0.8.0, which introduces belief-aware memory fields that prevent autonomous agent improvement loops from overwriting accurate information. The update adds `captureMode` and `beliefState` as fi…
Tim Berners-Lee, inventor of the World Wide Web, introduced Charlie, an AI assistant designed to protect user privacy by storing personal data in a user-controlled vault and requiring permission before sharing it with la…
OpenAI has released a policy document calling for mandatory pre-deployment safety evaluations of powerful AI models, a stricter stance than the Trump Administration's voluntary framework outlined in a recent executive or…
Data sovereignty has become a board-level AI risk in 2026 as regulated industries face new exposure from AI workflows that process, embed, and route data across multiple uncontrolled surfaces. The European Commission's 2…
Anthropic published a candid post-mortem on two years of deploying Claude as an autonomous agent, detailing security challenges and containment strategies, while also announcing a partnership with KPMG to deploy Claude t…
Snyk advises organizations to shift AI security budgets from fragmented tool spending to unified investments in visibility, governance, and control across the full AI lifecycle. The company warns that visibility without …
A new approach to secure AI code generation uses type-level security to make entire classes of web vulnerabilities uncompilable, inspired by Rust's memory safety guarantees and the Trusted Types API. The technique codifi…
A self-spreading npm worm called Miasma v2 compromised 57 packages across 286+ malicious versions by using a malicious binding.gyp file to trigger arbitrary code execution during npm install, without declaring lifecycle …
Enterprise AI teams are discovering that agentic AI systems face six critical production challenges—memory state, thundering herd, agent identity, blast radius, observability, and economics—that stem from distributed sys…
A new service called Liable Humans for AI provides organizations with verified human involvement at key decision points in AI workflows, including identity verification, authorized approvals, and document execution. The …