Coinbase and Binance Race to Access Anthropic's Mythos as DeFi Loses $292 Million to Infrastructure Exploits

Coinbase’s second-quarter outlook, published this week with data partner Glassnode, identifies Anthropic’s Mythos model as a direct threat to cryptocurrency markets. The same week, Aave’s community is still piecing together a $300 million recovery effort after the Kelp DAO exploit drained roughly $292 million through a bridge vulnerability on April 18. These two data points belong in the same sentence.

The crypto industry lost $3.4 billion to hacks and attacks in 2025, according to Chainalysis. Its primary defensive tool: smart contract audits that scan code before deployment and catalog known vulnerability patterns. Mythos does something different. It chains together weaknesses across systems, testing how protocols interact and how small flaws compound into real-world exploits. The audit model was built for code-level bugs. The exploits that are actually draining hundreds of millions target infrastructure: key management, bridges, oracle networks, signing services.

The Scramble for Access

Coinbase and Binance have both approached Anthropic to test Mythos on their own systems, The Information reported in April. Banks like JP Morgan are treating AI-driven cyber risk as systemic and running stress tests with similar tools, according to Bloomberg. Anthropic has not set a public release date for Mythos, citing concerns about misuse, and the model remains restricted to a small group of partners.

This creates an asymmetry. Organizations with access to Mythos can run adversarial simulations against their own infrastructure. Everyone else relies on the same audit playbook that missed the Kelp DAO bridge flaw, the Vercel breach that exposed crypto companies’ API keys through a third-party AI tool, and the $1 billion Hyperbridge attack that exploited cross-chain message verification.

“The bigger risks sit in infrastructure,” Paul Vijender, head of security at risk management firm Gauntlet, told CoinDesk. “When I think about AI-driven threats, I’m less concerned about smart contract exploits and more focused on AI-assisted attacks against the human and infrastructure layers.”

What Smart Contract Audits Miss

DeFi’s composability, the ability for protocols to connect, share liquidity, and build on each other, is both its defining feature and its most dangerous attack surface. A vulnerability in one protocol’s oracle feed or bridge implementation can cascade across every protocol that integrates with it. Mapping these dependencies at scale is beyond human audit capacity.

“Composability is what makes DeFi capital efficient and innovative,” Vijender said. “But it also means a minor vulnerability in one protocol can become a critical exploit vector with contagion potential across the ecosystem.”

The Kelp DAO exploit demonstrated this directly. Attackers exploited Kelp’s LayerZero-powered bridge to mint roughly 116,500 unbacked rsETH tokens, according to CoinDesk. The damage cascaded into Aave’s lending pools. By late April, Aave had raised roughly $160 million of the $200 million needed to cover bad debt, with Consensys, Lido, EtherFi, and others pledging support. Aave’s DAO allocated 250,000 ETH and founder Stani Kulechov donated 5,000 ETH personally.

Kulechov frames AI security tools as an evolution, not a disruption. “Web3 is no stranger to well-funded and motivated adversaries,” he told CoinDesk. “DeFi operates at compute speed, so AI doesn’t introduce a new dynamic. It intensifies an environment that has always required constant vigilance.”

The Widening Gap

Deddy Lavid, head of cybersecurity firm Cyvers Alert, told DL News that if AI can find weaknesses across the internet at scale, crypto could be one of the first sectors hit, because the industry relies on browsers, wallets, and open-source tools directly linked to moving money.

Uniswap Labs founder Hayden Adams offered a more measured view. “We haven’t tested Mythos yet, but we’re genuinely interested in what it and tools like it can do for protocol security,” he told CoinDesk. “AI gives builders better ways to stress test and harden systems.”

Adams expects the gap between secure and insecure protocols to widen: “Projects that prioritize security will have greater ability to test and harden systems before launching. Projects that don’t will be most at risk.”

That gap is the real story. Coinbase’s Q2 outlook notes that three-quarters of institutional investors still see Bitcoin as undervalued, and David Duong, Coinbase’s global head of investment research, says much of the speculative froth has already been flushed out. The market is not panicking over Mythos. But market sentiment and infrastructure security are different problems. The institutions that survive the next wave of AI-accelerated exploits will be the ones that had access to frontier testing tools before the exploits landed. The rest will be writing $300 million recovery proposals.