Author: DLNews

Translation: Deep Tide TechFlow

Deep Tide Overview: Anthropic's new AI model Claude Mythos can autonomously discover and exploit zero-day vulnerabilities; the company deems it too dangerous to sell publicly and only grants access to a few large tech partners. Coinbase and Binance are competing to purchase it to bolster defenses, but this means others in the industry will be left fully exposed to AI attacks. By 2025, the cryptocurrency industry had already lost 3.4 billion dollars due to hackers, while AI can find vulnerabilities completely overlooked by traditional audits in just seconds.

Major cryptocurrency exchanges are racing to gain access to Anthropic's Claude Mythos, a new AI model that is restricted to a few large tech partners due to its alleged hacking capabilities.

According to The Information on April 14, Coinbase and Binance are competing to purchase this new model to enhance defenses.

The urgency stems from Claude Mythos's alleged ability to autonomously discover and exploit zero-day vulnerabilities—security flaws that attackers can exploit even before developers know they exist—across major operating systems and web browsers. Worse yet, according to Anthropic, the model is too dangerous for the company to set a public release date.

However, as the largest exchanges rush to arm themselves with AI-driven defenses, others in the crypto space are left out—and vulnerable.

For Deddy Lavid, CEO of cybersecurity company Cyvers Alert, this scenario is nightmarish.

"If AI can identify vulnerabilities in core internet infrastructure on a large scale, crypto would be one of the first markets to feel the impact," Lavid told DL News.

"Our ecosystem operates on browsers, wallets, APIs, open-source libraries, and development tools—many of which are directly tied to value transfer." The potential damage is staggering.

"From hundreds of millions to tens of billions," he said.

According to blockchain monitoring company Chainalysis, the cryptocurrency industry had already suffered 3.4 billion dollars in losses due to hackers and malicious actors by 2025.

Indeed, for an industry built on irreversible transactions and streamlined security teams, the risks are existential. This is because AI does not only read code faster than humans; it quickly explores large amounts of code to find emerging vulnerabilities fully overlooked by traditional audits.

Combining this capability with the instantaneous finality and fragmented infrastructure of crypto shortens the gap between discovering vulnerabilities and catastrophic losses from weeks to seconds.

No layer is immune

Data from Cyvers shows that massive historical losses have hit every layer of the crypto trading stack.

Centralized exchanges lost nearly 2 billion dollars, decentralized networks were attacked with losses exceeding 500 million dollars, cross-chain bridges were breached for nearly 1 billion dollars, and user wallets became targets, Levid said.

AI threats could severely impact crypto in three main ways: small teams, long-unupdated codebases, or abandoned projects.

"We often see protocols worth millions being maintained by lean teams, aging codebases, or partially abandoned projects," Levid said. "In an AI-driven threat environment, these gaps become major risk amplifiers."

So what about Bitcoin, the dominant protocol in crypto?

"The core protocol of Bitcoin is relatively simple, time-tested, and much less exposed to complex logical errors than other ecosystems," Levid told DL News. "AI is unlikely to suddenly 'crack' Bitcoin itself."

But the infrastructure surrounding this top cryptocurrency—wallets, exchanges, L2—is still fragile.

"AI can discover vulnerabilities at these layers, where most users interact with Bitcoin, and actual losses will occur—not in the protocol itself, but at the edges," Levid said.