AI安全指数

About the Organization:The Future of Life Institute (FLI) is an independentnonprofit organization with the goal of reducing large-scale risks and steeringtransformative technologies to benefit humanity, with a particular focus onartificial intelligence (AI). Learn more at futureoflife.org.Contents1.2 Improvement opportunities by company1.4 Independent review panel3.2 Index Design and Structure3.3 Related Work and Incorporated Research3.4 Data Sources and Evidence Collection3.5 Grading Process and Expert Review4.2 Improvement opportunities by companyGovernance & AccountabilityWhistleblowing Policies (15 Questions)External Pre-Deployment Safety Testing (6 Questions)Internal Deployments (3 Questions)Safety Practices, Frameworks, and Teams (9 Questions) 1.1 Key Findings231.3 Methodology452Introduction63Methodology73.1 Companies Assessed771010113.6 Limitations114Results134.1 Key Findings13144.3 Domain-level findings155Conclusions20Appendix A: Grading Sheets21Risk Assessment22Current Harms33Safety Frameworks41Existential Safety4859Information Sharing71Appendix B: Company Survey85Introduction8586919495 21Executive SummaryThe Future of Life Institute's AI Safety Index provides an independent assessment of seven leading AI companies'efforts to manage both immediate harms and catastrophic risks from advanced AI systems. Conducted withan expert review panel of distinguished AI researchers and governance specialists, this second evaluationreveals an industry struggling to keep pace with its own rapid capability advances—with critical gaps in riskmanagement and safety planning that threaten our ability to control increasingly powerful AI systems.Grading:Uses the US GPA system for grade boundaries: A+, A, A-, B+, [...], F letter values corresponding to numerical values 4.3, 4.0, 3.7, 3.3, [...], 0.1.1 Key Findings•Anthropic gets the best overall grade (C+).The firm led on risk assessments, conducting the only humanparticipant bio-risk trials, excelled in privacy by not training on user data, conducted world-leading alignmentresearch, delivered strong safety benchmark performance, and demonstrated governance commitmentthrough its Public Benefit Corporation structure and proactive risk communication.•OpenAI secured second place ahead of Google DeepMind.OpenAI distinguished itself as the onlycompany to publish its whistleblowing policy, outlined a more robust risk management approach in itssafety framework, and assessed risks on pre-mitigation models. The company also shared more detailson external model evaluations, provided a detailed model specification, regularly disclosed instances ofmalicious misuse, and engaged comprehensively with the AI Safety Index survey.•The industry is fundamentally unprepared for its own stated goals.Companies claim they will achieveartificial general intelligence (AGI) within the decade, yet none scored above D in Existential Safety planning. Onereviewer called this disconnect “deeply disturbing,” noting that despite racing toward human-level AI, “none of thecompanies has anything like a coherent, actionable plan” for ensuring such systems remain safe and controllable.•Only 3 of 7 firms report substantive testing for dangerous capabilities linked to large-scale risks suchas bio- or cyber-terrorism(Anthropic, OpenAI, and Google DeepMind). While these leaders marginallyimproved the quality of their model cards, one reviewer warns that the underlying safety tests still missbasic risk-assessment standards: “The methodology/reasoning explicitly linking a given evaluation orexperimental procedure to the risk, with limitations and qualifications, is usually absent. [...] I have veryAnthropicOpenAIGoogleDeepMindOverallGradeC+CC-OverallScore2.642.101.76Risk AssessmentC+CC-Current HarmsB-BC+Safety FrameworksCCD+Existential SafetyDFD-Governance &AccountabilityA-C-DInformation SharingA-BB x.AIMetaZhipu AIDeepSeekDDFF1.231.060.620.37FDFFD+D+DDD+D+FFFFFFC-D-D+D+C+DDF 3low confidence that dangerous capabilities are being detected in time to prevent significant harm. Minimaloverall investment in external 3rd party evaluations decreases my confidence further.”•Capabilities are accelerating faster than risk management practice, and the gap between firms iswidening. With no common regulatory floor, a few motivated companies adopt stronger controls whileothers neglect basic safeguards, highlighting the inadequacy of voluntary pledges.•Whistleblowing policy transparency remains a weak spot.Public whistleblowing policies are a commonbest practice in safety-critical industries because they enable external scrutiny. Yet, among the assessedcompanies, only OpenAI has published its full policy, and it did so only after media reports revealed thepolicy’s highly restrictive non-disparagement clauses.•Chinese AI firms Zhipu.AI and DeepSeek both received failing overall grades.However, the report scorescompanies on norms such as self-governance and information-sharing, which are far less prominent inChinese corporate c