Crypto firms are still seeking frontier AI access, only a select few have it
Major crypto firms still lack access to top-tier restricted AI models meant to improve cybersecurity, while a few companies have secured early access. Coinbase has access to Anthropic’s gated Mythos model, and Anthropic reportedly used it to audit Zcash’s protocol. Binance, Fireblocks, and others say they have not obtained it. Anthropic and OpenAI both use tiered access systems that limit cyber-capable models to vetted defenders. Crypto security leaders generally support temporary safeguards, but argue restrictions should loosen as public models catch up and attackers can already use open-source tools. They say defenders need faster approval and broader access to keep pace with AI-assisted exploits. Some crypto-adjacent firms, including FIS and HackerOne, have joined Anthropic’s Project Glasswing for early access. At the same time, AI-assisted attacks appear to be increasing: Boltz halted a bridge after seeing more AI-driven exploits, and Coinkite said a wallet flaw may have been found with AI.
