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.
Inbox Zero, an open-source AI email client with over 28,000 GitHub stars, implements a three-tier prompt injection defense system to protect against adversarial email content. The system, detailed in a dedicated `securit…
Blackstone and Guggenheim are reducing software loan allocations in new collateralized loan obligations due to concerns that artificial intelligence will disrupt the business models of software companies. The two major C…
Linux kernel maintainer Greg Kroah-Hartman told the Rust Week conference that the Rust programming language will save Linux from a surge of security vulnerabilities discovered by AI-powered bug detection tools. Kroah-Har…
On March 17, 2025, Elon Musk warned during a podcast with US Senator Ted Cruz that America's reliance on Taiwan for advanced AI chip manufacturing poses a critical national security risk if China invades the island. Musk…
The OpenAI Foundation committed $250 million on May 27 to address economic disruption caused by AI-driven labor displacement. The funding will support grants, partnerships, and initiatives to measure AI's economic impact…
The OpenAI Foundation is committing an initial $250 million to grants, partnerships, and direct work aimed at building secure and abundant economic futures as artificial intelligence drives major economic changes. The pr…
A developer has open-sourced a public-safe demo of "execution-boundary governance" for AI coding agents, a system that focuses on what a model is allowed to propose before any action is taken. The repository demonstrates…
Google Research introduced a new private analytics system combining cryptographic secure aggregation with trusted execution environments (TEEs) to achieve zero-trust privacy guarantees. The solution processes data locall…
A16z crypto researchers tested AI agents against 20 real Ethereum price manipulation cases, finding the agents could detect vulnerabilities in all cases but only successfully execute exploits in 10% of baseline tests. Wh…
Anthropic's Model Psych team published three papers exploring how large language models can introspect on their own emotional states, finding that models like Claude activate emotion vectors that influence behavior simil…
Robinhood launched a beta program called Agentic Trading on May 27, allowing users to connect third-party AI agents to dedicated accounts for stock trading. The retail trading giant plans to expand the feature to crypto,…
Geodesic Research, a Cambridge, UK-based non-profit AI safety organization, announced its mission to develop robust alignment initializations for capable large language models, focusing on preventing misalignment that ar…
A 2026 Cloudian research report found that 75% of 212 senior IT decision makers had moved workloads from the cloud back to on-premises infrastructure in the prior 24 months, driven by security, compliance, AI workloads, …
A developer released an AI-agent skill that audits software supply chains for compromised packages, designed to detect exposure to supply-chain attacks. The tool is scheduled to run twice weekly to provide ongoing monito…
A developer created piighost-proofreader, a tool that anonymizes CVs locally before sending them to an LLM for proofreading, ensuring no personal data like names, addresses, or employer details are exposed to the third-p…
Palo Alto Networks CEO Nikesh Arora warns that the next six months in cybersecurity will be unprecedented, with AI serving as both the primary threat and the only viable defense. In a conversation with South Park Commons…
Huawei researchers, in collaboration with the Beijing Institute of Technology and Peking University, released the Claw-Anything benchmark on May 25, 2026, revealing that top AI models like GPT-5.5 and Claude Opus 4.7 sco…
OpenAI and Anthropic have spent millions in a political proxy war over AI regulation, but their efforts to defeat New York state assemblyman Alex Bores have backfired, turning the once-obscure candidate into a front-runn…
Researchers found that all major AI chatbots failed compliance tests designed to assess adherence to the European Union's upcoming AI regulations. The study revealed that none of the prominent bots met the required stand…
A single-character bug in the Starlette open-source framework, tracked as CVE-2026-48710 and nicknamed BadHost, allowed attackers to bypass authentication on millions of servers running AI agents and MCP tools by injecti…