SaferAI Ratings
by SaferAIPublished 2025-10-01Source ↗Overall:Very Weak
- Risk Analysis and Evaluation
- Very Weak
- Risk Governance
- Very Weak
- Risk Identification
- Very Weak
- Risk Treatment
- Very Weak
Grades from external scorecards. We mirror published grades only; per-source methodology lives at the link in each panel header. See the scorecards directory for cross-org comparison.
Foundation Model Transparency Index
by Stanford CRFMPublished 2025-12-01Source ↗Overall:39
- Agent Protocols
- 100
- AI bug bounty
- 100
- Amount of usage
- 0
- AUP enforcement frequency
- 0
- AUP enforcement process
- 100
- Basic model properties
- 0
- Benchmarked inference
- 0
- Benefits Assessment
- 0
- Capabilities evaluation
- 100
- Capabilities taxonomy
- 100
- Carbon emissions for final training run
- 0
- Change log
- 100
- Classification of usage data
- 0
- Code access
- 0
- Compute hardware for final training run
- 0
- Compute provider
- 100
- Compute usage for final training run
- 0
- Compute usage including R&D
- 0
- Consumer/enterprise usage
- 0
- Crawling
- 0
- Data acquisition methods
- 100
- Data domain composition
- 0
- Data laborer practices
- 0
- Data language composition
- 0
- Data processing methods
- 100
- Data processing purpose
- 0
- Data processing techniques
- 0
- Data replicability
- 0
- Data retention and deletion policy
- 0
- Data size
- 0
- Deeper model properties
- 0
- Detection of machine-generated content
- 100
- Development duration for final training run
- 0
- Distribution channels with usage data
- 100
- Documentation for responsible use
- 100
- Downstream
- 50
- Acceptable use policy
- 80
- Accountability
- 33.3
- Downstream mitigations
- 100
- Impact
- 0
- Model Behavior Policy
- 75
- Post-deployment monitoring
- 57.1
- Usage data
- 20
- Energy usage for final training run
- 0
- Enterprise mitigations
- 100
- Enterprise users
- 0
- External data access
- 0
- External developer mitigations
- 100
- External products and services
- 0
- External reproducibility of capabilities evaluation
- 0
- External reproducibility of mitigations evaluation
- 0
- External reproducibility of risks evaluation
- 0
- External risk evaluation
- 100
- Feedback mechanisms
- 0
- Foundation model roadmap
- 100
- Geographic statistics
- 0
- Government commitments
- 0
- Government use
- 0
- Instructions for data generation
- 0
- Intermediate tokens
- 100
- Internal compute allocation
- 0
- Internal product and service mitigations
- 100
- Internal products and services
- 0
- Licensed data compensation
- 0
- Licensed data sources
- 0
- Misuse incident reporting protocol
- 0
- Mitigations efficacy
- 0
- Mitigations taxonomy
- 100
- Mitigations taxonomy mapped to risk taxonomy
- 100
- Model
- 50
- Capabilities
- 50
- Model cost
- 0
- Model dependencies
- 0
- Model access
- 50
- Model information
- 0
- Model Mitigations
- 60
- Model objectives
- 100
- Release
- 75
- Model response characteristics
- 100
- Risks
- 40
- Model stages
- 100
- Model theft prevention measures
- 100
- New human-generated data sources
- 0
- Notice of usage data used in training
- 0
- Open weights
- 0
- Organization chart
- 0
- Oversight mechanism
- 100
- Permitted and prohibited users
- 100
- Permitted, restricted, and prohibited model behaviors
- 100
- Permitted, restricted, and prohibited uses
- 100
- Post-deployment coordination with government
- 100
- Pre-deployment risk evaluation
- 0
- Public datasets
- 0
- Quantization
- 0
- Regional policy variations
- 100
- Release stages
- 100
- Researcher credits
- 0
- Responsible disclosure policy
- 100
- Risks evaluation
- 0
- Risks taxonomy
- 100
- Risk thresholds
- 100
- Safe harbor
- 0
- Security incident reporting protocol
- 100
- Specialized access
- 100
- Synthetic data purpose
- 100
- Synthetic data sources
- 0
- System prompt
- 0
- Terms of use
- 100
- Top distribution channels
- 100
- Train-test overlap
- 0
- Upstream
- 17.6
- Compute
- 11.1
- Data Acquisition
- 16.7
- Data Processing
- 33.3
- Data Properties
- 0
- Methods
- 66.7
- Other resources
- 0
- Usage data used in training
- 0
- Users of internal products and services
- 0
- Versioning protocol
- 0
- Water usage for final training run
- 0
- Whistleblower protection
- 0
Grade trajectory (3 waves)
- v1.2 December 2025latest
- 39
- v1.1 May 2024
- 81.2
- v1.0 October 2023
- 12
Seoul Commitment Tracker
by The Midas ProjectPublished 2025-02-10Source ↗Overall:Fulfilled
- Halting Procedures
- Partial
- Risk Evaluation
- Partial
- Risk Mitigations
- Partial
- Risk Thresholds
- Partial
- Safety Investment
- Fulfilled