Anthropic opens its strongest hacking-capable models to many more security teams

The post is dated Oct. 6, 2026. Anthropic says each tier includes its most capable models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models after those.

Who can apply for Defense Access, in Anthropic’s list: company, nonprofit, university, and government security teams; operators of critical infrastructure of any size, “such as regional hospitals or municipal utilities”; smaller security firms; open-source maintainers; and individual researchers with a track record. Anthropic says it aims to answer within a few days.

Red Team review takes “a few weeks,” and only organizations qualify. Even then, Anthropic says real-time blocks stay on for things such as deploying ransomware or damaging physical systems. Specialized Access is where Project Glasswing members move. Anthropic says it reviews every organization in that tier in depth with the U.S. government. The post does not name which agency.

Members must let Anthropic keep their data so it can watch for misuse, until a planned “Enterprise Frontier Safeguards” option arrives “later this fall.”

Anthropic’s own test of Claude Opus 5.5 on a hacking benchmark it calls CyScenarioBench, 50 trials: with no program access, every task was blocked on the first prompt. In Defense Access, 46 of 50 were blocked at some point. In Red Team Access, none were blocked, and 34 of 50 succeeded. Those figures are Anthropic’s.

Anthropic also says Glasswing partners found “at least 129,000 verified software vulnerabilities” from April to July 2026, and that its own scanning found 5,500 more from April to October. More than 33,000 were rated critical or high severity. Anthropic says this is based on survey data from only some partners, and that the true impact is likely “at least five times higher.” That is a company estimate, not a count this desk checked.