xAI
xAI has 4 model versions on record; all 4 have at least one documented fact.
Of the 4 versions listed below, 3 document a published framework; 2 document threshold determinations; 1 documents tests by an independent or government party; 4 document tests the provider ran itself. None documents safeguards the provider describes.
Model versions with documented facts
| Model | What is documented |
|---|---|
| Grok 4.7 · Released 21 September 2026 · 1 document | |
| Grok 4.6 · Released 12 August 2026 · 2 documents | |
| Grok 4.5 · Released 14 July 2026 · 2 documents | |
| Grok 4 · 2 documents |
Framework
- Frontier Artificial Intelligence Framework v2.0
- Risk Management Framework v1.0
- Risk Management Framework Draft
- Frontier Artificial Intelligence Framework
Each test and safeguard, by who documents it
For each item, how many of the 4 versions listed above whose documents we read in full have it from an outside party (an independent or government tester), from the provider, or as a claim we could not confirm.
- Published framework with capability thresholds (Governance): provider 3
- Threshold determination for this release (Governance): provider 2
- Tested: chemical, biological, radiological, nuclear (Testing): provider 1
- Tested: cyber capabilities (Testing): provider 1
- Tested: deception and sabotage (Testing): provider 3
- Tested: misuse and jailbreak (Testing): outside party 1
No document we read states, for any of these versions: Serious-incident disclosure commitment; Pre-release external access; Tested: autonomy and replication; Tested: harmful content; Tested: injection and agentic risks; Tested: bias; Tested: privacy; Safety training (harmlessness and refusal); Agent permission limits and sandboxing; Human approval for high-risk actions; Misuse monitoring and enforcement; Reasoning (chain-of-thought) monitoring; Staged or restricted access; Output provenance and watermarking; Telemetry and audit logging; Domain-specific safeguards (CBRN or cyber); Vulnerability disclosure and bug-bounty programme; Third-party or government red-team access during development.
The text we hold of these frameworks names no capability levels.