- Published on
Crypto firms still seeking frontier AI access; only select few have it
- Authors
- Name
- DeepHorus
Crypto firms are still waiting for access to the most advanced frontier AI models, with only a handful of industry players securing the powerful tools needed to defend against sophisticated cyberattacks. US exchange Coinbase obtained access to Anthropic's restricted Mythos model in June, and Zcash's Zooko Wilcox confirmed Anthropic used it to audit the Zcash protocol. But many other major crypto firms, including Binance, remain locked out, creating a new security divide in an industry where a single exploit can cost billions.
The uneven access comes at a critical time. Blockchain networks hold billions in assets, and attackers are increasingly using AI to find vulnerabilities and automate exploits. While model developers like Anthropic and OpenAI restrict their most cyber-capable models to vetted defenders, open-source alternatives are rapidly improving. Crypto executives worry that defenders will be left with weaker tools while attackers can leverage capable open models, widening the security gap.
Binance chief security officer Jimmy Su told Cointelegraph that the exchange has been trying to gain access to Anthropic's most advanced model but has not succeeded. "We have been trying to make inroads there. We also talked to other crypto exchanges and our own investors to try to make some progress," Su said. He acknowledged that only a select few have received the cutting-edge model, leaving most major players to rely on less powerful defenses.
Technically, these restricted models are highly capable but dangerous. Anthropic's Mythos 5 uses the same underlying model as its public Fable 5, but without safety guardrails that block sensitive cybersecurity work. OpenAI operates a similar tiered system: verified defenders can use GPT-5.5 with "Trusted Access for Cyber," while the more permissive GPT-5.5-Cyber is reserved for a smaller group doing authorized penetration testing. This tiering is designed to prevent misuse while still supporting legitimate security research.
The short-term impact is clear: crypto firms without access face higher risk from AI-assisted attacks. In the long term, the balance may shift. As open-source models approach frontier capabilities, restricting access becomes harder to justify. Su noted that "as other more powerful models are being released, the pressure will be on Anthropic to make it more widely available." The question is whether defenders can deploy the model as effectively as attackers once it becomes public.
Industry reactions are mixed but lean toward cautious support for restrictions with a push for faster access. Su called Anthropic's controlled rollout responsible, because a new model might benefit attackers faster than defenders, increasing the "blast radius." However, Solana Foundation CISO Michael Coates argued that legitimate defenders need a streamlined path: "We need to streamline the verification programs, the acceptance programs, to give these models to legitimate defenders." Blockchain Capital's Sean Cheetham went further, saying broader availability would favor defenders since legitimate researchers vastly outnumber sophisticated attackers.
Looking ahead, expect growing pressure on AI developers to open access as open-source models improve. Crypto security teams will likely continue lobbying for inclusion, while Anthropic and OpenAI refine their vetting processes. The outcome will shape whether the industry can close the security divide or leave the most vulnerable projects exposed to AI-driven attacks in the coming years.
