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

Back to watch register
11Information integrityBelow the top 20

Memory-holing via model updates

CapabilityDomain knowledgeImpact domainBoth

Statement (NASA form)

Given that what a popular model will say about a person or event changes with each release, there is a possibility of an update that quietly drops, softens, or invents a public fact resulting in a reference layer the public treats as a library and that is not one.

Likelihood
3Probable
Consequence
3Major
Urgency
2Planned

A library does not silently reprint last year’s encyclopaedia.

Composite 11 = 3×3 + 2

Applicable mitigations

Related on the map