Silent wrong state or drift
The system reports success, or keeps running, while its output or the state it left is wrong.
NIST AI 600-1 · No exact NIST AI 600-1 categoryMAS AI risk management guidelines (2026)NIST AI RMF MEASURE 2.4
2incidents (primary) · 2 more as secondary
0read for controls · 2 not yet
0fixes reported
0repeats after a fix
Incidents per month
No rise: 1 in Oct 2026, against an average of 0.3 over the 3 months before.
Developers
Anthropic · 1
Where the chain broke
| Part of the system | Incidents where the control was not effective or did not exist | Incidents where the control was effective | Incidents where sources don't mention the control | Incidents not yet checked for controls |
|---|---|---|---|---|
| L1 Model Model provider | 2 | |||
| L2 Harness Agent framework vendor | 2 | |||
| L3 Application Deployer | 2 | |||
| L4 Infrastructure / target Organisation hit | 2 |
Controls for this risk, by layer
| Control | Incidents where the control was not effective or did not exist | Incidents where the control was effective | Incidents where sources don't mention the control | Most common reason it was not effective |
|---|---|---|---|---|
| Change review and separation of duties L3 Application | ||||
| Activity logs L3 Application | ||||
| Live behaviour monitoring L3 Application | ||||
| Drift monitoring L3 Application | ||||
| Independent check of task results L3 Application | ||||
| Grounding and fact checks L3 Application |
6 of 6 controls for this risk not yet recorded as effective.
Fixes reported and what followed
No remediation reported for incidents of this risk.
Incidents
EffectiveNot effectiveNot effective (bypassed)Did not existNot stated in sourcesNot yet checkedAttacker's ownOrder: model · harness · application · target
| Date | Incident | Risk | Where it failed | Controls mentioned in sources |
|---|---|---|---|---|
| 2026-10-29 | LLM agents in chemistry lab-reasoning study ignore experimental evidence and assert unsupported claims | Silent wrong state or drift | Not yet checked | |
| 2026-09-01 | Claude Code Auto Mode: website summary leads to remote code execution via Python module shadowing Anthropic |
Silent wrong state or drift | Not yet checked |
Also under this risk as a secondary risk: LR-2026-0603, LR-2026-0481