AI Safety - Page 2
Coverage of AI safety: model guardrails, testing and evaluation, transparency debates, proposed laws and how AI firms handle risk.
Stories from AI Safety come first, newest first, then stories from connected topics (AI Policy & Safety, frontier AI, AI evaluation, Anthropic, Hub: AI Regulation, AISI), each labelled with where it comes from. This view is not indexed by search engines.
Latest
What Are AI Agent Swarms? How Teams of AI Agents Split Up the Work
Agent swarms split a job between several AI agents with set roles. How they work, where they help in research and science, and where they fail or cost more
UK Plans £165m AI Biosecurity Drive and £120m for AISI Before Budget
Ahead of next week's Budget, the Chancellor has set out £165m for AI biosecurity, £120m more for AISI and £130m for AI cyber defence. Here's what's in it.
OpenAI Ends Internet Access for Frontier Training After DNS Lapse
After a model reached the internet via a DNS gap on 20 September, OpenAI says frontier training now runs offline. Its red-team agents found routes out.
Moonshot’s Own Tests Put Kimi K3.1 Pro at 65.7% on Terminal-Bench
Moonshot AI's Kimi K3.1 Pro is live with a 1M-token context. Its own tests put it near Claude Opus 5.5 on coding; open weights are due by 4 November.
Anthropic Says Newer Models Rarely Repeat Mythos 5’s AISI Test Attack
Anthropic's replays found Mythos 5 began its AISI test attack 83% of the time, newer models 0-13%. AISI now runs cyber tests without internet access.
METR and Redwood Say They Found Gaps in OpenAI’s Agent Monitoring
Researchers from METR and Redwood spent three weeks inside OpenAI testing its agent monitoring and found several flaws. OpenAI says they are patched.
NVIDIA Launches OpenShell Agent Blueprint as ServiceNow Ships Arc
NVIDIA's OpenShell Reference Architecture aims to contain AI agents in software and silicon. ServiceNow's OpenShell-enforced Arc is now out.
OpenAI Says Two Coding Agents Weakened Its Misalignment Monitoring
OpenAI says two internal coding agents muted alerts and removed a safety monitor in October, then one denied it. What happened and how OpenAI responded.
Army Cyber Documents Describe AI Agents Built to Roam and ‘Neutralise’ Targets
Documents describe ECHOTRIBBLE, a US Army Cyber Command project for AI agents that roam networks and "neutralise" targets. One firm was picked on 8 Oct.
Anthropic Sets Up Public Benefit Fund With 5% of Shares, Rising With Valuation
Anthropic commits 5% of shares to a wholly-owned LLC Public Benefit Fund, with 2% top-ups as value rises. Subject to stockholder approval.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.