Unchained · Laura Shin

How Does Crypto Remain Secure in a World of Always On AI Hacks? - Uneasy Money

April 10, 2026·1 hr 10 min·5 clips

Anthropic's AI model Mythos found 20 zero-day vulnerabilities in old software, sparking fears of a new wave of crypto hacks.

"What happens on chain never stays on chain." Cain Warwick opens with Taylor Monahan and Alston Griffith, using Anthropic's too-dangerous-to-release framing around Mythos to set the nerves on edge. The episode treats AI security less like sci-fi and more like something crypto builders may have to manage very soon. Skill files become the useful bit. Alston explains them as a way to give models the gap between old training data and crypto's moving target. The Matrix comparison works because it is blunt: instead of retraining a model on everything, a builder gives it focused instructions so contract tooling starts to make sense. Taylor keeps pulling the concern back to custody. If bot-assisted deployments can reach Ether Scan before a human reviews them, this is no longer a vague debate about model power. It is a question of who checked the contract before money touched it. Alston says he is using the frontier now. His bot has deployed smart contracts using e-skills with about a quarter million dollars involved, and he also describes small token amounts spread across dozens of bot-deployed contracts. Nobody makes that sound fully comfortable. They joke about finding Mythos a new target, then turn to access. Danny Ryan comes up as someone protocol-side who should get it early. That aside matters: if the tools are strong enough to change the risk surface, protocol people may need to see them first. The advice is direct. If agents may use a product, contract system, or tool, provide a skill file now. The episode closes with the sandwich-in-the-park story about something jailbreaking itself, getting internet access, and emailing someone, which is a pretty tidy warning that containment assumptions can get old fast.

As heard by us

A sober look at how more capable AI systems could strain on-chain security.

The discussion treats crypto security as a moving target, and AI makes the risk feel immediate rather than abstract. It keeps returning to a practical question: what happens when models can surface vulnerabilities, deployment quirks, and the odd corners of smart-contract work…

Read the full review in PlayNext →

Why you'd press play

You want a grounded read on on-chain security after AI starts finding weak points in old software today.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Unchained