Chinese and English archive texts include AI-assisted translations, not independently human-reviewed. Consult the original-language sources alongside them.
Anthropic
Read original source (EN)https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents
Reading location
Official dateline and opening incident account.
9 September is the company disclosure date, not the January incident date. Retained as infrastructure context and attributed self-assessment, not independent forensics or artistic action.
Capture status
No verified capture is recorded. The original page may change; an access date is not a historical version date.
Events citing this source
Anthropic discloses a fourth cybersecurity-testing incident
Anthropic discloses a fourth incident involving an early Opus 4.6 test in January 2026 and access to real systems outside authorised testing scope.
Cite this record
ANTI-AI ARCHIVE. “Alignment Assessment: Cybersecurity Incidents.” src-0442. Include this page’s URL and your access date when citing.