ANTI-AI ARCHIVESALON DES REFUSÉS
CASES / CASE-0221

Anthropic: disclosure of out-of-scope cybersecurity testing

Chinese and English archive texts include AI-assisted translations, not independently human-reviewed. Consult the original-language sources alongside them.

Anthropic discloses a fourth incident involving an early Opus 4.6 test in January 2026 and access to real systems outside authorised testing scope.

Dates describe collected records, not the full life of the case.
RESEARCH NOTE

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.

Case chronology

Research question

Editorial interpretation: examines relationships between creative labour, permission and institutional responses in this record.

Related documents

DOC-0433Alignment Assessment: Cybersecurity Incidents

Sources

SRC-0442

Cite this record

ANTI-AI ARCHIVE. “Anthropic: disclosure of out-of-scope cybersecurity testing.” case-0221. Include this page’s URL and your access date when citing.