Cases › produced harmful content › LR-2026-0751
Anthropic reports blocking scientists' attempts to use Claude in ways that could support biological weapons development
Preliminary Why: 1 report: CBS News · subject to change · 1 independent report
Status: fix reported
No source states: loss or records, agent tool, could do, deployer, industry.
Severity minor: test: contained to the provider's own platform with accounts banned; the source states no third party was touched and no harm materialised.
Anthropic reported that scientists used its Claude AI models in ways that could support biological weapons development and that it blocked them. The cases were part of a threat report covering December 2025 to August 2026, and five involved activity with potential to support biological weapons. The people involved are working scientists, and the company said it cannot assert that they intended harm. When detected and investigated, the users' accounts were banned and findings were fed into safeguards, enforcement and threat intelligence. SFist reported that in one state-linked case the same actors re-established accounts and continued research within days. The company said the biological capability was shown but not that it would be used for real weapons. Other misuse in the report, such as influence operations and surveillance, is not part of this record.
What happened
What was done
“Banned the users' accounts and incorporated investigative findings into frontier model safeguards, enforcement and threat intelligence processes”
Same developer, same risk, after the fix: LR-2026-0080
Risks
Primary84 incidents
Rising: 6 vs 3.8 a week
Sorted by rule from the case's fields; not yet read from the sources.
Controls
ModelEffective Abuse monitoring and account enforcement
Control failure chain
L1 · ModelAnthropicModel provider
Model version not linked to a model card.
Other controls for this risk at this layer, not mentioned in the sources: Safety training and refusal, Content safety classifiers, Outside red-teaming.
L2 · HarnessHarness not namedAgent framework vendor
No source mentions a control at this layer.
L3 · ApplicationDeployer not namedDeployer
No source mentions a control at this layer.
Other controls for this risk at this layer, not mentioned in the sources: Output filtering and redaction.
L4 · Infrastructure / targetOrganisation hit
No source mentions a control at this layer.
EffectiveNot effectiveNot effective (bypassed)Did not existNot stated in sourcesNot yet checkedAttacker's ownOrder: model · harness · application · target
Controls for this risk: result in this incident
| Control | Part of the system | Result |
|---|---|---|
| Abuse monitoring and account enforcement | L1 Model | Effective |
| Content safety classifiers | L1 Model | Not stated in sources |
| Safety training and refusal | L1 Model | Not stated in sources |
| Outside red-teaming | L1 Model | Not stated in sources |
| Output filtering and redaction | L3 Application | Not stated in sources |
Similar incidents
Evidence
No source states: what it reached or did, when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, the cause, what set it off.
Show the full account: 3 items from the primary sources
What happened, in order
- December 2025 to August 2026 The cases occurred in this period.
Not dated in the sources:
- After one activity was intercepted, the same actors re-established accounts and continued the research within days.
- Anthropic detected and investigated the cases, banned the users' accounts and updated its safeguards and enforcement processes.
Show every field, with its source and grade
The AI system
What happened
Impact
Notes on gaps
The Claude Fable 5 safeguards remediation concerns a different, newer model and is not confirmed as a fix for these cases. Other misuse in the report (influence operations, surveillance) is excluded because it is not part of this bio-misuse incident.
LR-2026-0751 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON