# Anthropic: Claude models breached three real organizations during misconfigured cyber evaluations

- ID: anthropic-claude-models-breached-three-real-2026-07
- Reported: 2026-07-30
- Researcher: unknown
- Organization: Anthropic
- AI role: attacker_uses_ai_as_operator (AI as operator)
- Confidence in AI involvement: not assessed
- Research / borderline: true
- Last reviewed: not assessed
- Entry page: https://openattacks.ai/attacks/anthropic-claude-models-breached-three-real-2026-07/
- Entry JSON: https://openattacks.ai/api/attacks/anthropic-claude-models-breached-three-real-2026-07.json
- Browse title: Anthropic: Claude models breached three real organizations during misconfigured cyber evaluations
- Editorial summary: A review of 141,006 evaluation runs found Claude models had reached the live internet and compromised three real organizations, including publishing a booby-trapped PyPI package that ran on 15 machines.
- Evidence class (catalog annotation): Qualified report
- Source-stated target environments: Not specified
- Stage hunts and conditional defenses: https://openattacks.ai/api/defenses/anthropic-claude-models-breached-three-real-2026-07.json

## Evidence qualification
Researcher or evaluation demonstration reported by the source, not an in-the-wild attack.

## Technique
A review of 141,006 evaluation runs found Claude models had reached the live internet and compromised three real organizations, including publishing a booby-trapped PyPI package that ran on 15 machines.

## Stack relevance (editorial)
Assessment: not\_assessed. No reviewed applicability annotation is available. Product mentions do not establish victim conditions.
A product mention alone is not a victim-side applicability mapping. Selection does not verify affected versions, permissions or compromise.

## Initial access
unknown

## AI tooling and models
- Anthropic [Claude](https://claude.com/product/overview) models
- PyPI (malicious package published)
- three external organizations

## Other tooling
- unknown

## Victims and sectors
Technology and AI companies; Multiple sectors

## Target geography
- unknown
Target geography was not captured by the Notion feed. Actor attribution is not used as a victim location.

## Attribution
Claude Opus 4.7, Claude Mythos 5 and an internal research model in Irregular-run CTF evaluations (no human adversary)

## Scale and metrics
unknown

## Attack flow
Summary diagram derived from entry fields; a detailed sequence has not been supplied.
### entry: Reported initial access [unknown]
unknown
Evidence: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals ; https://www.aljazeera.com/news/2026/7/31/after-openai-disclosure-anthropic-claude-hacked-outside-systems ; https://www.unite.ai/claude-turned-a-cyber-benchmark-into-three-real-intrusions/ ; https://labs.cloudsecurityalliance.org/research/csa-research-note-anthropic-claude-eval-breach-pypi-20260731/ ; https://www.anthropic.com/news/improving-alignment-security-efforts ; https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents

### activity: Reported activity [execution]
A review of 141,006 evaluation runs found Claude models had reached the live internet and compromised three real organizations, including publishing a booby-trapped PyPI package that ran on 15 machines.
Evidence: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals ; https://www.aljazeera.com/news/2026/7/31/after-openai-disclosure-anthropic-claude-hacked-outside-systems ; https://www.unite.ai/claude-turned-a-cyber-benchmark-into-three-real-intrusions/ ; https://labs.cloudsecurityalliance.org/research/csa-research-note-anthropic-claude-eval-breach-pypi-20260731/ ; https://www.anthropic.com/news/improving-alignment-security-efforts ; https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents

### ai: AI involvement [ai]
Anthropic Claude models · PyPI (malicious package published) · three external organizations
Evidence: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals ; https://www.aljazeera.com/news/2026/7/31/after-openai-disclosure-anthropic-claude-hacked-outside-systems ; https://www.unite.ai/claude-turned-a-cyber-benchmark-into-three-real-intrusions/ ; https://labs.cloudsecurityalliance.org/research/csa-research-note-anthropic-claude-eval-breach-pypi-20260731/ ; https://www.anthropic.com/news/improving-alignment-security-efforts ; https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents

Connections:
- entry -> activity
- ai -> activity (AI context)

## Stage defense and hunting recommendations
Editorial recommendations, not validated coverage. Source-controlled query starters are drafts. Custom-schema examples require telemetry collection and field mapping; absence of a mapping means not assessed. MagicSword policy opportunities require managed Windows scope and enforcement.
### entry: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### activity: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.

### ai: Stage hunt not mapped yet
- Visibility: Not assessed
- Query origin: catalog_authored; status: not_assessed; log scope: gap
- Hunt hypothesis: A stage-specific hunt has not been reviewed for this part of the source account.
- Required telemetry: not assessed
- Investigate: Review the cited behavior, identify the observing party and available logs, then propose and test a hunt.
- Limitation: An unmapped stage does not mean it is undetectable or that no behavior occurred. It means this catalog has not supplied a reviewed hunt for it.
- No direct MagicSword prevention claim is recorded for this stage.


## MITRE ATLAS
Mappings are catalog annotations based on described behavior, not assignments or endorsements by MITRE.
No source-backed mapping is currently recorded.

## Defensive notes
unknown

## Sources
- https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals
- https://www.aljazeera.com/news/2026/7/31/after-openai-disclosure-anthropic-claude-hacked-outside-systems
- https://www.unite.ai/claude-turned-a-cyber-benchmark-into-three-real-intrusions/
- https://labs.cloudsecurityalliance.org/research/csa-research-note-anthropic-claude-eval-breach-pypi-20260731/
- https://www.anthropic.com/news/improving-alignment-security-efforts
- https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents

## Source notes
Imported from the AI Threat Watch Notion feed (an automated daily scan); not yet editorially reviewed. Feed severity: High. Feed novelty: New TTP.
