Coinbase and Binance are seeking access to Anthropic’s Mythos model as crypto exchanges and custodians move to prepare for a new wave of AI driven cyber threats, according to a report from The Information.
The report said Coinbase is in close communication with Anthropic about Mythos, with Coinbase CSO Philip Martin saying the model will accelerate digital threats as well as defense, while Binance and Fireblocks are also taking steps to understand how far it could reshape cyber offense and defense tools.
The push comes just one week after Anthropic introduced Claude Mythos Preview, a new frontier model the company described as unusually capable at cybersecurity tasks. Anthropic said the model can identify and exploit zero-day vulnerabilities across major operating systems and web browsers.
Mythos is not being released broadly. Instead, Anthropic is limiting access to a small group of partner organizations through Project Glasswing, a new initiative that provides early access to select companies including Amazon Web Services, Apple, Google, JPMorganChase, Microsoft, Nvidia, Palo Alto Networks, and others.
Anthropic acknowledged in late March that it was testing a new model representing a step change in performance after a data leak exposed its existence. The leak involved a draft blog post about the model, then internally referred to as Capybara, that had been left in an unsecured cache. That was separate from Anthropic’s early April Claude Code source leak, which exposed internal code for its coding tool rather than the Mythos model itself.
Fireblocks had already been testing Anthropic’s earlier Opus 4.6 model, which Anthropic said was able to find meaningful zero-day vulnerabilities in well tested codebases without specialized scaffolding.
In February, Anthropic said Opus 4.6 had found more than 500 previously unknown high severity flaws in open source libraries, underscoring why crypto infrastructure firms are paying attention to what comes next with Mythos.
