What happened
Future of Life Institute's semi-annual AI Safety Index grades nine leading AI companies (Anthropic, OpenAI, Google DeepMind, Meta, Z.ai, Alibaba Cloud, xAI, DeepSeek, Mistral) across six domains — Risk Assessment, Current Harms, Safety Frameworks, Existential Safety, Governance & Accountability, and Information Sharing — using an independent expert review panel and a US GPA-style grading system (A+ to F). Anthropic again earns the top overall grade (C+, 2.66), while three companies — xAI, DeepSeek, and Mistral (Europe's top AI company) — receive outright failing grades, with the report noting 'inadequate safety is a global problem, not a regional one.' The methodology draws on company surveys, published safety frameworks, and third-party benchmarks (Stanford HELM, TrustLLM, CAIS); the report also flags that several leading labs — including Anthropic, OpenAI, Google DeepMind, and Meta — have 'weakened or voided' prior pause/redline commitments, which reviewers call a 'moving goalpost' undermining safety frameworks industry-wide. Existential Safety is identified as the weakest domain industry-wide, with no company scoring above a C-.
Why it matters
Boards and CISOs evaluating AI vendor risk, procurement, or partnership decisions now have an independent, cross-company comparative safety scorecard with a specific verbatim finding: safety commitments are eroding even among industry leaders, and no company adequately addresses existential/catastrophic risk domains.
Action needed
Map current and prospective AI vendor relationships against the Index's per-company grades and specifically flag any reliance on labs graded D or below in Existential Safety and Governance domains.