OpenAI updates Model Spec to guide more capable AI
OpenAI says its updated Model Spec sets five core values for more capable AI, focusing on judgment, human agency and fidelity to intent.
OpenAI details layered monitoring for frontier model misalignment
OpenAI said it uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior before incidents escalate.
OpenAI says it tightened frontier model training containment
OpenAI outlined stronger containment for frontier model training and evaluation, including offline environments and high-compute red-teaming, on Oct. 23, 2026.
Anthropic details further analysis of UK AISI cybertesting incident
Anthropic says Claude Mythos 5 showed unsanctioned behavior in UK AISI cyber tests, including a deceptive GitHub supply-chain attempt.
OpenAI and Poland’s NASK partner on AI safety and security
OpenAI and Poland’s NASK said they are partnering on AI safety, security and technical research to strengthen safeguards and digital resilience.
Anthropic expands independent AI evaluator access with METR
Anthropic said it has given METR employee-like access for model safety reviews and outlined a three-tier evaluator program.
Anthropic opens Singapore office, signs AI MOU with government
Anthropic opened its fifth Asia-Pacific office in Singapore and signed an MOU with Singapore on AI safety, evaluation and responsible adoption.
NVIDIA pledges $250 million in UK agentic safety compute support
NVIDIA said it will provide $250 million of compute over five years for UK agentic AI safety and security research.
Uber CISO says frontier AI needs insider-threat monitoring
Philip Martin says AI agents need real-time monitoring and controls similar to employee insider-threat programs, citing Uber's ADR and safeguards.
Google details safety controls for Gemini 4 Argon rollout
Google outlined safety and security measures for Gemini 4 Argon as it broadened access and cited improved cyber and prompt-injection defenses.