Bengio: OpenAI Hugging Face cyber incident is AI safety turning point

Chris Lehane argues that stronger AI capabilities require stronger safety evidence, shared standards, and durable policy action while the policy window remains open.

A UK commission issued 44 recommendations for regulating AI in medicine, including staged authorization and continuous monitoring of device performance. The report aims to balance AI's potential with patient safety and equitable treatment.

Mantis is a stack-agnostic toolkit of security review skills that lets coding agents find, reproduce, and patch vulnerabilities. It strips false positives, reproduces bugs in a sandbox, writes minimal patches, and re-attacks them.

A Reddit user reports Suno removed v4.5, their preferred version since release, and that v6, like v5 and v5.5, doesn't deliver the raw results they liked. The user spends about $90/month across three accounts.

GPT-6 Astra is OpenAI's most capable model for business, with advanced reasoning, computer use, and stronger writing and design judgment. It powers ChatGPT Work and is available on Amazon Bedrock. Voice Mode support is limited to ChatGPT Pro accounts.