AI Will Make the Next 12 Months 'a Slaughterhouse' for Crypto Security: Anchorage Exec
Key Takeaways
- •AI systems have advanced within the past three to six months from simply finding vulnerabilities to autonomously building exploits and carrying out every step of an attack, enabling anyone without security expertise to launch hacks.
- •North Korean hackers have stolen more than $13 billion in cryptocurrency over the past decade, including over $7.5 billion this year and roughly $6 billion from last month's Binance hack, which the FBI attributed to North Korea.
- •North Korean operatives commonly use fake job interviews, poisoned software downloads, and deepfake video calls, with deepfake calls alone accounting for $400 million in stolen funds.
- •The Solana Foundation's free STRIDE security framework has been completed by protocols covering more than 74% of DeFi total value locked on Solana, and its defensive AI reverse-engineering effort prevented over $5 million in potential losses in the past month.
- •Anchorage Digital now onboards staff in person and equips employees with locked-down Chromebooks connected to remote development machines, a setup the company says removes 90% to 95% of its corporate attack surface.

Crypto companies are bracing for a wave of AI-powered attacks over the coming year, layered on top of what security leaders describe as an already relentless threat from North Korea. The warnings came at Breakpoint, Solana's annual conference, held this year in London.
"The next 12 months are going to be absolutely brutal," said Diogo Mónica, co-founder of crypto bank Anchorage Digital. "They're going to be a slaughterhouse."
Over the past three to six months, AI systems have advanced from simply finding vulnerabilities to building full exploits and carrying out every step of an attack on their own, Mónica said. Speaking on a panel chaired by Gordon Corera, a former BBC security correspondent and co-host of The Rest Is Classified podcast, he added that aggressive hacking tools now let anyone with no security knowledge simply tell a model to "please hack this website."
Michael Coates, chief information security officer at the Solana Foundation, the nonprofit that supports the Solana blockchain, said AI has "dramatically lowered the cost and the barrier to hacking." Projects that once escaped attention will now be targeted opportunistically, he said.
'Part of their GDP'
North Korea remains the biggest threat to crypto firms, Coates said in a separate presentation. More than $13 billion in crypto has been stolen over the past decade, over $7.5 billion of it this year alone, with some $6 billion drained in a single breach: last month's hack of Binance, which the FBI has blamed on North Korea.
"This is how they fund their programs in North Korea," he said.
Coates described the Binance attack as a chain of familiar tactics. A developer downloaded poisoned software from a sponsored search result, and an engineer later took a job interview on a work device, downloading software as part of it. That gave the attackers a foothold to move through the network and reach signing services, which they used to authorize malicious transactions rather than stealing the keys themselves, he said.
Fake job interviews are a "telltale tactic of North Korea," he said, along with poisoned tools and deepfake video calls, in which AI-generated faces and voices impersonate real people. North Korean hackers have stolen $400 million using deepfake calls alone. When North Korean operatives can't get hired directly, they use people in the U.S. to take jobs on their behalf, then connect remote keyboards and mice so attackers can access the company's network, Mónica said.
The U.S. Treasury recently sanctioned 11 people over just such a North Korean IT worker scheme. Anchorage now onboards new staff in person at its U.S. office, Mónica added, and every employee works on a locked-down Chromebook, logging into remote machines for development—a setup he said removes 90% to 95% of the company's corporate attack surface, the full set of entry points an intruder could try to exploit.
'Protect the keys'
Coates pointed teams to STRIDE, a free security framework the Solana Foundation launched in the spring, borrowing its name from the classic threat-modeling categories of spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege. It sets out 18 core controls tailored for crypto, and protocols covering more than 74% of DeFi's total value locked on Solana—a standard yardstick for the funds deposited across decentralized finance protocols—including 10 of the top 11, have gone through it, he said. The foundation is also using AI defensively, reverse-engineering newly launched programs to find flaws before attackers do—an effort he said has prevented more than $5 million in potential losses in the past month.
Both executives warned against basic mistakes. Connecting email inboxes to AI agents invites prompt injection attacks, in which hidden or crafted instructions trick an AI model into unintended actions, Mónica said, while Coates urged teams to "protect the keys," using hardware security keys for logins and properly configured multisig wallets, which require multiple separate keys to approve a single transaction, with plans for key rotation and loss.
"That's where we see the attackers going," Coates said. "They're getting the entry points through the basics. They're pivoting around to get the keys, and that's where they're unleashing their fury."