Anthropic has restricted access to its newest frontier model, Mythos, due to its heightened ability to detect and exploit software vulnerabilities across critical systems. Rather than releasing the model publicly, the company is providing it to a select group of organisations managing essential infrastructure, including Amazon Web Services and JPMorgan Chase, with a focus on defensive cybersecurity applications.
The company indicated that Mythos significantly surpasses earlier models such as Opus in identifying exploitable weaknesses, highlighting the accelerating capabilities of AI in cybersecurity. This controlled deployment reflects growing concerns about the misuse of advanced models and the need to safeguard sensitive systems.
The strategy also reflects a broader shift among AI labs toward limiting access to top-tier models, prioritising enterprise partnerships while addressing both security challenges and competitive pressures in the AI ecosystem.