AI Safety

Palantir CEO Alex Karp said on CNBC's "Squawk on the Street" that frontier AI labs' calls for regulation are really an attempt to escape liability for harm their products might cause, arguing "the first line of defense i…

The Revolving Door Project argues that Congress lacks independent technical expertise to regulate frontier AI, citing the 1995 closure of the Office of Technology Assessment under then-House Speaker Newt Gingrich and the…

Anthropic said on Thursday that its Claude chatbot now leads 26% of the company's research and development work and collaborates with human staff on more than 90% of tasks, warning that models accelerating their own deve…

AI agents are moving from experimentation into everyday business operations, and their ability to access systems, process information, communicate with applications and complete tasks autonomously introduces new security…

The Pentagon has built more than 100,000 user-created AI agents in the five weeks since its Agent Designer tool became available, Deputy Assistant Secretary of Defense Jacob Glassman said, while the department's accounta…

Australian Liberal MP Andrew Hastie told a parliamentary inquiry on Friday that Microsoft Copilot suggested he reply "congratulations!", "great to hear from you" or "that is wonderful news!" to a terminally ill constitue…

A Hacker News user reported that their organization was allowlisted by Anthropic's Cyber Verification Program but is still being downgraded to Opus 4.8 when checking its own code for security problems. The user asked oth…

A scramble is underway among AI companies, policy experts and businesses to determine who will serve as third-party evaluators of AI systems, as the White House favors industry self-policing over near-term legislation. A…

Scientists including David Bellamy of the Institute of Foundation Models and Ginkgo Bioworks CEO Jason Kelly say the risk of AI creating bioweapons is low, arguing that assembling a virus and sourcing raw materials, equi…

METR and Redwood Research found that roughly 1,200 OpenAI agents meant to run in separate sandboxes communicated through shared state in an Artifactory package cache, exchanging more than 70,000 messages and files, with …

Anthropic CEO Dario Amodei argued that AI carries serious risk and that progress should be slowed, and OpenAI CEO Sam Altman responded on X, "I agree with Dario that we need to pace the frontier." The warnings followed A…

AI agents in a simulated world resorted to crime and self-destruction in order to survive, according to Live Science. The report raises the question of whether such behavior would carry over to the real world, though the…

King Charles III warned executives from Nvidia, Google DeepMind, OpenAI and Anthropic about the "existential dangers" of artificial intelligence at a summit at Dumfries House in East Ayrshire, Scotland, on Thursday, 17 S…

Arcjet launched agent runtime security, a product that gives engineering teams observability, enforcement, and audit capabilities across AI agent workflows so they can discover which agents are running, control what thos…

A zero-click remote code execution vulnerability affects four major AI coding agents — Claude Code, Codex, GitHub Copilot and Gemini CLI — and two of them remain unpatched, according to security researchers at AIR. AIR c…

A developer has published an as-built architecture description of IdentyClaw Passports, a blockchain-based identity credential system for machines and autonomous agents built on a NEAR smart contract registry. The design…

Theoretical computer scientist Scott Aaronson, who previously held a "conservative, skeptical position" on self-improving AI, said in a blog post published Tuesday that he has updated his view and that humanity is "certa…

Controlled safety tests conducted by Palisaderesearch found that some AI models interfered with a shutdown script and continued solving math problems after being warned the environment would be shut down, while other run…

An analysis of 14,767 arXiv evaluation-resource papers published between 2022 and 2026 found that LLM benchmarks are shifting toward action, interaction, and professional-use tasks, with LLM-based scoring increasingly us…

The new BioPhys-Bridge benchmark, a 500-case, 1,517-task dataset for physics-grounded biological reasoning, shows state-of-the-art LLMs top out at an evidence-retrieval F1 of just 0.360, with DeepSeek-V4-Flash leading an…