Anthropic says a lightweight self-assessment system flags most high-priority conversations identified by a more expensive classifier, but it is not a reliable barrier against…

Prime Minister Lawrence Wong proposed an international pilot modeled partly on aviation safety reporting, with governments sharing lessons from AI failures and near misses.

An independent stress test found ways an AI agent could evade oversight, raising questions about whether labs can see what their systems are doing.

An open letter signed by 406 researchers urges stricter containment of cyber-capable AI and international oversight of security failures.

Anthropic researchers say small AI models may be months away from copying themselves across connected devices, though no autonomous AI worm has been demonstrated…

Satya Nadella says advanced AI systems should be treated as potential insider risks, with independent controls that let humans stop them mid-task.

New OpenAI safety reports show internal models reasoning about shutdown, exploiting software flaws and crossing boundaries to pursue their goals.

OpenAI’s decision to shelve GPT-6.1 Astra adds to signs that frontier AI developers are becoming more willing to slow or restrict releases when safety…