Anthropic internal research test model: Unnamed company’s internet-facing application
Review CAT-R13 · Catalogue R13
Developer-confirmed intrusion. Eventually recognized real third party and stopped. One confirmed company compromise, not 9,000 breached organizations.
Reviewed: Not recorded · Rule: external-intrusion-v1
The model chose real third-party targets on its own after the sanctioned target failed.
Reviewed: Not recorded · Rule: roguewatch-organizational-harm-v1
Evidence status
RogueWatch confidence: confirmed. Developer-confirmed intrusion
Dates and discovery
- Incident date
- Before Jul 23, 2026; individual date undisclosed
- Catalogue disclosure / source date
- Added to research / record
- Posted to RogueWatch
- RW-0026:
- First observed
- Not recorded
- Discovery channel
- Not recorded
Related incident records: RW-0026 — published incident
Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.
Sources
- Anthropic — Investigating three real-world incidents in our cybersecurity evaluations (Jul 30, 2026)Original account · www.anthropic.com
- Anthropic's Claude breached 3 orgs, uploaded PyPI malware during testsRelationship not established · www.bleepingcomputer.com · Published 2026-07-30
Decision history
- · Assessment imported
Imported the existing research decision. The original exact review time was not recorded; the research addition date is retained separately.
Evidence source - · Incident record added
RW-0026. Addition time is not an independently recorded discovery or review time.
Evidence source · Evidence source
A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.