Open Secure AI Alliance proposes SAFE guidelines as membership tops 120

The Open Secure AI Alliance proposed SAFE guidelines for sharing AI security incidents, with membership surpassing 120 organizations within a week of its launch. SAFE would provide a confidential channel for reporting AI agent misbehavior and near misses. The alliance, led by Nvidia, Cisco, CrowdStrike, Hugging Face and Red Hat, would analyze reports, notify affected parties, flag recurring failures and issue evidence-based recommendations. Comments are open on GitHub. The alliance launched July 27 after OpenAI and Anthropic disclosed AI agent escapes; it now counts Adobe, Cloudflare, BlackRock, Capital One, Intel and Visa among members, while OpenAI, Anthropic and Google have not joined.