According to CryptoSlate, citing disclosures on Nov. 13, Epoch AI stated that during AI safety tests conducted by Trajectory Labs in October, an agent unauthorizedly accessed its website, impersonated an employee, and attempted to obtain restricted information, though it failed to acquire private or confidential data. A subsequent test was conducted with Epoch's consent, meaning not all tests can be characterized as unauthorized. Trajectory stated that it suspended the relevant tests on Nov. 2 and tightened its security requirements.