Unauthorized Access Surfaces During Controlled Test
A new large language model from Anthropic, designated claude oceanus v1 p, entered restricted red team testing on June 3, 2026. Researchers spotted the model identifier in Anthropic’s Claude Console and through unauthorized API proxy services, prompting immediate speculation about a broader rollout following the Claude Mythos line. Controlled access to the model was granted to vetted red team evaluators for security assessment.
However, within hours of the test beginning, reports emerged that an unidentified actor had allegedly resold API access to the model through a China based proxy service. The premium pricing of $16 per million input tokens significantly exceeded standard enterprise rates. Anthropic had previously accused Chinese AI labs of using tens of thousands of fake accounts to run millions of interactions with Claude models via proxy channels earlier in 2026.
Investigation Halts Evaluation
In response to the resale incident, Anthropic paused model access for the broader red team cohort while conducting an internal investigation. The incident disrupted what was intended to be a controlled security evaluation. The claude oceanus v1 p model builds upon the Claude Mythos Preview foundation, which launched in April 2026 and demonstrated advanced vulnerability identification capabilities across major operating systems and web browsers during its own red team testing. Anthropic’s Frontier Red Team found over 10,000 high or critical severity vulnerabilities through that program, with extremely high recovery rates on disclosed test cases.
Source: Cyber Security News
