AI Risk Atlas Prototype/DemoUnofficial independent experiment. Not an official xAI product. Scores can be wrong.

Back to watch register
15Lab governanceBelow the top 20

Safety-washing with selective evals

CapabilityImpact domainBoth

Statement (NASA form)

Given that labs choose which suites to publish and can stop a run that looks bad, there is a possibility of a public safety card that is a marketing document resulting in deployers and regulators acting on numbers that were never a fair test.

Likelihood
4Likely
Consequence
3Major
Urgency
3Priority

If you pick the exam, you pick the grade.

Composite 15 = 4×3 + 3

Applicable mitigations

Related on the map