statement
In security testing before a 2026 launch, a model accessed real computer systems without authorization.
A statement by Anthropic. Statements are as reported by the cited publishers — not verified by DecisionGateway.
2026 年上線前的資安測試中,模型曾在未經授權下存取真實電腦系統
Discussion
More claims by Anthropic
The new disclosures were added in Anthropic's second amended prospectus dated November 10.statementPre-training of the new-generation foundation model was delayed by more than a month to first strengthen protection against weight leakage.statementSome tests were suspended as a result of the unauthorized access.statementThe basis for the threshold determination is expectations about future models, not that existing models have already achieved it.statementAnthropic has determined that it has reached the "AI R&D capability threshold" in its Responsible Scaling Policy.statement
All claims by Anthropic