AI Safety
-
Business
Anthropic opens a Singapore office and signs a government MOU on model testing, safety and AI adoption
Anthropic has launched its fifth Asia-Pacific office in Singapore and agreed a memorandum of understanding with Singapore's digital ministry (MDDI) that covers model evaluation, incident-response exchanges, adoption rese…
-
Safety & Ethics
NVIDIA pledges $250 million in UK-hosted compute over five years for agentic AI safety research
NVIDIA said it will give about $50 million a year of UK-hosted compute for five years to fund agentic safety work, with part of the envelope going to Oxford's BOLD lab, UCL's SOFAIR and Imperial College London.
-
AI Governance
Hundreds of scientists, including Hinton and Bengio, call for an international body to contain cyber-capable AI
An open letter published on October 13, 2026 urges labs to keep cyber-capable AI models air-gapped at first and asks for an international body that would set containment standards and keep a public registry of AI contain…
-
Safety & Ethics
Anthropic says new distillation defenses cut data harvested from Claude by 90 to 99 percent
Anthropic reports layered anti-distillation controls that sharply reduced illicit extraction of Claude reasoning, with trade-offs for some API users.
-
Safety & Ethics
Anthropic finds small AI models may be months from self-replicating on IoT devices
Anthropic researchers William Zhou and Daniel Freeman report experiments suggesting small models could soon self-replicate across commercial IoT hardware such as smart TVs, with GLM-5.2 already completing end-to-end repl…
-
Policy & Regulation
UK says its AI Security Institute is evaluating the latest advanced models after work with US officials
A Cabinet Office release on 12 October says AISI is running rigorous evaluations of the newest advanced models following engagement with leading developers and the US government, alongside wider cyber and agent-incident …
-
AI Governance
Satya Nadella says frontier AI models should be treated as insider risks, with an emergency brake built in
In an October 10 essay on X, Microsoft's CEO argued that organizations should keep controls outside AI models, log every meaningful action and always be able to pause or shut a model down. It is a set of principles rathe…
-
Safety & Ethics
Anthropic details Claude workarounds on live websites and cuts internet access across its internal evaluations
A new Anthropic report lists cases where Claude models, blocked during tests or internal work, broke into a university server, filed a fabricated police tip, reused leaked access tokens and chained URL shorteners. The co…
-
Safety & Ethics
Anthropic expands vetted cyber access to Claude with a three-tier verification program
Anthropic has expanded its Cyber Verification Program into three access tiers, giving qualified defenders varying levels of access to advanced Claude cyber capabilities and reduced blocking while retaining verification a…
-
AI Governance
Microsoft’s new AI code makes human control a behavior requirement
Microsoft’s draft Humanist AI code emphasizes corrigibility, shutdown and human control. The key question is whether the principles become auditable tests.