Anthropic is preparing a new model called Claude Mythos, according to details exposed in a data leak reported by Fortune on March 26. While Anthropic did not confirm the leaked blog post to AI Business, it confirmed to other publications that it has completed training on a model called Mythos. The leaked document described Mythos as Anthropic’s most powerful and compute-intensive model and said it is advanced at identifying and exploiting software vulnerabilities, raising both commercial interest and cybersecurity concerns.
Mythos’ cybersecurity focus triggered a sharp decline in cybersecurity stocks on March 27, reflecting investor anxiety that foundation model providers could encroach on established software categories. The reaction echoed responses in other sectors after Anthropic introduced Claude Cowork in February. Analysts see the move as part of Anthropic’s effort to diversify beyond coding and pursue a wider set of Artificial Intelligence applications. Lian Jye Su of Omdia said a sole focus on coding had become limiting for Anthropic and that a targeted move into security shows the company is trying to broaden its market and present itself as a serious security partner.
Questions remain about Mythos’ actual performance, but expectations are that it could exceed Anthropic’s current models. Anthropic has already shown interest in the area with Claude Security Code, which it released in a limited preview in February. Even so, the market response appears overstated. Startups in cybersecurity may feel pressure from the growing capabilities of foundation model developers, but larger vendors such as Palo Alto Networks and CrowdStrike are less likely to see Mythos as an existential threat.
Enterprise buyers are still expected to rely on a mix of providers and tools. Many cybersecurity companies already build on models from foundation model vendors, with differentiation coming from fine-tuning and internal development. For businesses, the bigger takeaway is that Artificial Intelligence is becoming more deeply embedded across applications. In cybersecurity environments shaped by shadow Artificial Intelligence, distributed systems from cloud to edge, and fragmented data estates, organizations will need multiple tools and partners, with Anthropic serving as one option rather than the only one.
