Anthropic's Claude Mythos has emerged as the world's most powerful AI model for identifying and exploiting cybersecurity vulnerabilities, triggering a geopolitical standoff over who gets access to it. China has expressed anxiety about the model being weaponized, while major cryptocurrency companies like Binance remain locked out of the restricted tool despite holding billions in assets.

What Makes Claude Mythos So Powerful and Dangerous?

Claude Mythos represents a significant leap in AI capabilities for cybersecurity work. The model can rapidly identify security flaws in software, generate code to exploit them, and autonomously carry out attacks on computer systems. What makes it particularly concerning is that it has nearly eliminated the time gap between when a vulnerability is discovered and when it can be exploited, leaving cybersecurity professionals with minimal opportunity to patch the flaw before attackers strike.

The Trump administration recognized the dual-use nature of this technology and blocked Anthropic from providing Mythos and its companion model, Fable, to foreign nationals or organizations on June 12. However, the administration lifted restrictions on Fable access on June 30, though Mythos remains tightly controlled.

Why Is China So Concerned About Claude Mythos?

Chinese officials have raised concerns about Anthropic's Mythos and Fable models not being made available to Chinese companies and researchers, according to Bloomberg reporting. While China is not preparing immediate action against Anthropic, government officials are drawing up potential responses should the Trump administration escalate technology restrictions further. These contingency plans include potential sanctions or outright bans on American AI companies.

The tension reflects a broader pattern of U.S. technology restrictions aimed at limiting China's access to advanced capabilities. The Trump administration has previously blocked China from accessing Nvidia's most advanced chips and recently limited imports of foreign-made robots and inverters. However, experts warn these restrictions could backfire. Chinese AI models, while less advanced than American alternatives, are cheaper and largely open-sourced, and their usage among American companies has surged dramatically. According to OpenRouter data, the share of tokens used by American companies on Chinese AI models has remained above 30 percent each week since February, compared to just 11 percent average over the past 12 months.

Which Crypto Companies Have Access to Frontier AI Models?

The uneven distribution of access to Claude Mythos has created a new security divide in the cryptocurrency industry, where exploits can put billions of dollars at risk. Only a select few crypto firms have secured access to the restricted model. Coinbase announced in June that it had obtained access to Mythos, and Zcash's Zooko Wilcox confirmed that Anthropic used the model to audit the Zcash protocol at the request of Shielded Labs. However, major players like Binance, the world's largest crypto exchange by daily trading volume and holder of $137.8 billion in assets, have been unable to gain approval.

"That's one advanced frontier model that hasn't been made available to crypto just yet. 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. But we haven't gotten the most frontier AI model, like Mythos," said Jimmy Su, Binance's chief security officer.

Jimmy Su, Chief Security Officer at Binance

Other major crypto infrastructure companies remain in limbo. Fireblocks, a crypto custodian that secures trillions in assets annually, sought access to Mythos in April but was still using only Anthropic's publicly available model for security testing at that time. Uniswap founder Hayden Adams criticized Fable 5's safeguards that restrict cybersecurity-related prompts. The Ethereum Foundation disclosed in July that it has been running coordinated AI agents to find bugs across its systems but did not specify which models were being used.

How Are Security Experts Weighing Restricted Access?

Anthropic's approach to controlling access to Mythos reflects a deliberate strategy to balance security concerns with practical defense needs. The company says Mythos 5 uses the same underlying model as its publicly available Fable 5 but without safeguards that restrict sensitive cybersecurity work. OpenAI operates a similar tiered system, with verified defenders able to use GPT-5.5 with "Trusted Access for Cyber," while a smaller group conducting authorized penetration testing can access the more permissive GPT-5.5-Cyber model.

Crypto security executives acknowledge the logic behind initial restrictions but argue the justification weakens as competing models become more capable. Binance's Su explained that restricted access makes sense when a newly released model benefits attackers faster than defenders, potentially harming the broader ecosystem. However, he noted that this calculation changes as other powerful models become available.

"If it enhances the attacker much faster than the defender, then it actually is harming the ecosystem. As other more powerful models are being released, the pressure will be on Anthropic to make it more widely available," noted Jimmy Su.

Jimmy Su, Chief Security Officer at Binance

Steps Organizations Can Take to Strengthen Cybersecurity in the AI Era

• Apply for Verified Access Programs: Organizations can apply to Anthropic's Project Glasswing, a gated program that provides vetted cyber defenders and organizations responsible for critical software infrastructure early access to restricted Mythos models, though approval is selective and competitive.
• Implement Coordinated AI Bug-Finding: Like the Ethereum Foundation, organizations can deploy coordinated AI agents to systematically identify vulnerabilities across their systems using publicly available models while seeking approval for more advanced tools.
• Establish Rapid Patching Protocols: Given that Mythos has nearly eliminated the time between vulnerability discovery and exploitation, organizations must develop streamlined processes to identify, test, and deploy security patches within hours rather than days.
• Diversify Security Testing Approaches: Organizations should combine multiple AI models and traditional security testing methods rather than relying on a single frontier model, reducing dependency on restricted access while maintaining defense capabilities.

What Real-World Damage Has AI-Assisted Exploitation Caused?

The risks posed by advanced AI cybersecurity tools are no longer theoretical. Bitcoin swap service Boltz announced on Monday that it halted its non-custodial bridge after observing a steady rise in AI-assisted exploits over recent months. The company stated that "the pattern is clear: attackers now iterate faster than a team our size can find and patch".

Bitcoin hardware wallet company Coinkite also reported that multiple Coldcard devices were exploited due to a flaw in its wallet seed generation, which proved less random than expected. Coinkite speculated that the attacker had used AI to review previous firmware versions to identify and exploit the vulnerability, despite the company having used "one of the best available AI models" to review its code just weeks before the attack.

These incidents underscore why access to frontier AI models like Claude Mythos has become so contested. Defenders argue they need the same powerful tools that attackers can access through open-source alternatives or illicit channels. Solana Foundation Chief Information Security Officer Michael Coates, who joined the foundation in July, emphasized the urgency of streamlining access for legitimate defenders.

"I fully understand guardrails for advanced models, but we need to streamline the verification programs, the acceptance programs, to give these models to legitimate defenders. We need to make sure that the best models we can get are in the hands of defenders because attackers will have something capable enough," explained Michael Coates.

Michael Coates, Chief Information Security Officer at Solana Foundation

The debate over Claude Mythos access reflects a fundamental tension in AI governance: how to enable beneficial uses of powerful technology while preventing misuse. As more sophisticated open-source alternatives emerge and AI-assisted attacks become more frequent, pressure will mount on Anthropic and other AI developers to expand access to frontier models beyond their current restricted circles. The outcome will likely shape cybersecurity practices across industries for years to come.