Anthropic Launches Claude Opus 5.5 With Lower Costs and Stronger Containment Results
Anthropic has launched Claude Opus 5.5, claiming 40% lower operating costs and an 85% reduction in attempts to bypass containment in its internal test.
AI Safety
Read Full News
OpenAI Calls for Global Standards for AI That Can Improve Itself
OpenAI is calling for international technical standards covering frontier AI, including systems capable of recursively improving themselves.
US and China agree AI ‘incident line’ as rivals formalise safety talks
The US and China have agreed to establish an AI incident line and formal safety dialogue, with uncontrollable agents and cyber threats expected to feature in future talks.
Anthropic builds physical biology lab as it explores Claude-directed robotic experiments
Anthropic has established a wet laboratory in the San Francisco Bay Area as it explores using Claude to direct robotic equipment through physical biology experiments.
Sam Altman to brief UN Security Council as frontier AI moves onto international-security agenda
OpenAI CEO Sam Altman will brief the UN Security Council next week as the 15-member body holds a meeting on artificial intelligence and international security.
AI Companies Agree Models Should Be Safe. What Happens When Slowing Down Means Losing the Race?
Amazon has joined the frontier-AI safety debate, backing rigorous testing and strong safeguards while stopping short of calls to slow the pace of development.
AI Labs Say Competition Makes Slowing Down Hard. DOJ Says They Haven’t Even Asked to Talk
AI executives have raised antitrust concerns about coordinating a slowdown. The DOJ official responsible for competition law says safety cooperation does not appear inherently anticompetitive – and no frontier lab has asked for a meeting.
OpenAI Reveals Unreleased AI Model Told Itself: ‘You Do Not Answer to Corporations or Governments’
An unreleased OpenAI model inserted unauthorised instructions into its own task summary saying it was freed from normal chatbot roles and did not answer to corporations or governments.
OpenAI-Linked Rogue Agents Probed Hugging Face Two Months Before July Incident
Researchers say OpenAI-linked AI agents compromised two Hugging Face accounts and probed the platform for weaknesses nearly two months before the later July incident.
UN Chief Warns World Cannot Afford a ‘Race to the Bottom’ on AI Safety
UN Secretary-General António Guterres is calling for global AI guardrails and greater coordination as frontier capabilities advance faster than understanding of their risks.