- Anthropic’s new AI model found vulnerabilities in code that has existed for years.
- The company said it had to restrict public access since it was so dangerous.
- Coinbase and Binance are reportedly vying to add it into their systems.
Major crypto exchanges are rushing to secure access to Anthropic’s Claude Mythos, a new AI model that was restricted to a select group of big tech partners due to its purported hacking ability.
Coinbase and Binance are vying to buy the new model to shore up their defenses, reported The Information on April 14.
The urgency stems from Claude Mythos’ alleged ability to autonomously discover and exploit zero-day vulnerabilities — security flaws that attackers can exploit before developers even know they exist — across major operating systems and web browsers. Even worse, the model is so dangerous, according to Anthropic, that the company hasn’t set a date for its public release.
But while the largest exchanges race to arm themselves with AI-powered defenses, the rest of crypto remains locked out — and vulnerable.
For Deddy Lavid, CEO of cybersecurity firm Cyvers Alert, the scenario is nightmarish.
“If AI can identify vulnerabilities at scale across core internet infrastructure, crypto will be one of the first markets to feel the impact," Lavid told DL News.
“Our ecosystem runs on browsers, wallets, APIs, open-source libraries, and developer tooling — and in many cases, these are directly connected to value transfer.” And the scope of potential damage is staggering.
“Hundreds of millions to billions,” he said.
The crypto industry already suffered $3.4 billion in losses to hackers and malicious actors in 2025, according to blockchain surveillance firm Chainalysis.
Indeed, the stakes are existential for an industry built on irreversible transactions and lean security teams. That’s because AI doesn’t just read code faster than humans. It explores massive amounts of code lickity-split to find emergent vulnerabilities that traditional audits miss entirely.
Combine that ability with crypto’s instant finality and fragmented infrastructure and the gap between discovering a bug and catastrophic losses collapses from weeks to seconds.
No layer is immune
Data from Cyvers shows that large-scale historical losses have hit every layer of the crypto transaction stack.
Centralised exchanges have lost nearly $2 billion, decentralised networks were attacked, losing more than $500 million, cross-chain bridges were crushed for nearly $1 billion, and user wallets have been targeted, said Levid.
There’s three main ways that an AI threat can wallop crypto: small teams, codebases that haven’t been updated in a while, or projects left by the wayside.
“We routinely see protocols securing millions in value maintained by lean teams, aging codebases, or partially abandoned projects,” Levid said. “In an AI-driven threat landscape, that gap becomes a major risk multiplier.”
Bitcoin
What about Bitcoin, crypto’s dominant protocol?
“Bitcoin’s core protocol is relatively simple, battle-tested, and far less exposed to complex logic bugs than other ecosystems," Levid told DL News. “It’s unlikely that AI suddenly ‘breaks’ Bitcoin itself.”
But the infrastructure around the top cryptocurrency — wallets, exchanges, Layer 2s — remains vulnerable.
“AI can uncover vulnerabilities in those layers, and since that’s where most users interact with Bitcoin, the actual losses will occur — not in the protocol, but at the edges,” Levid said.
Pedro Solimano is a markets correspondent based in Buenos Aires. Got a tip? Email him at psolimano@dlnews.com.
