Scorecard/Anthropic

Anthropic

A-

AI safety company building reliable, interpretable AI systems. Developer of Claude.

HQUS

Est2021

Size1001-5000

EU AI ActGPAI Systemic Risk

Score

72.8 / 100

Evidence

15 items

Industry-leading safety practices with comprehensive governance and technical safeguards.

Strengths:Governance Maturity, Technical Safety, Risk Assessment, Regulatory Readiness, External Engagement

Focus Areas

foundation modelsai safetyalignment researchenterprise ai

Safety Profile

Dimension Breakdown

Governance Maturitymedium

Published policies, corporate structure, safety mandate, whistleblowing, executive commitment.

8 evidence items

GM-01GM-02GM-03GM-04

Technical Safetymedium

Benchmarks, adversarial robustness, fine-tuning safety, watermarking, model cards, research output.

4 evidence items

TS-01TS-06TS-07

Risk Assessmentlow

Dangerous capability evaluations, thresholds, external testing, bug bounty, halt conditions.

1 evidence items

RA-01

Regulatory Readinesslow

ISO 42001, EU AI Act compliance, GPAI obligations, international commitments, incident reporting.

External Engagementmedium

Survey participation, research support, transparency, behavior specs, open-source contributions.

2 evidence items

EE-01EE-02

Indicator-level evidence, peer benchmarking, regulatory gap analysis, and quarterly monitoring.