BETA RELEASE

AI Safety Index, Winter 2025

Research by the Future of Life Institute. The most comprehensive independent assessment of AI company safety practices to date. Read the full report.

AI Safety Index Winter 2025 Scorecard from the Future of Life Institute. Shows safety grades for 8 major AI companies: Anthropic (C+), OpenAI (C+), Google DeepMind (C), xAI (D), Z.ai (D), Meta (D-), DeepSeek (D-), Alibaba Cloud (D-). Grades span six domains: Risk Assessment, Current Harms, Safety Frameworks, Existential Safety, Governance and Accountability, and Information Sharing.
Click to enlarge. Source: Future of Life Institute, Winter 2025.

The best overall grade is a C+.

No company scored above D on existential safety. Every company scored below C on at least one critical domain.

What this means for you

Risk Assessment
Identifying and evaluating risks before deploying models.
Before using an AI tool, look for information about known risks for your application.
Current Harms
Addressing bias, misuse, and misinformation in current products.
These harms are happening now, not hypothetically. See real incidents.
Safety Frameworks
Structured safety processes, red-teaming, and testing protocols.
Look for third-party audits, not self-assessments. A company's safety page is marketing until someone else verifies it.
Existential Safety
Plans for catastrophic or irreversible AI risks.
No company scored above a D here. The hardest problems are getting the least attention. If this concerns you, talk about it. Awareness is the first step toward accountability.
Governance and Accountability
Independent oversight, not just internal review.
Check whether your AI provider has an independent safety board.
Information Sharing
Publishing safety research and incident reports.
Compare what companies publish about their safety testing. The ones sharing the least usually have the most to hide.