Priority shifts toward visibility
Security expert Taylor Monahan says crypto teams should prioritize monitoring and automatic shutoffs over repeated audits because her advice changed over the last six months as AI agents began to probe projects more frequently. Monahan’s advice to teams used to be more classic advice, but she now emphasizes visibility above all else. Monahan argues that since audits provide 90% coverage and formal verification adds 5%, teams still face a 5% chance of complete collapse that requires an autonomous circuit breaker to pull the plug. AI allows small teams to watch systems for anomalies, which replaces work that once required highly skilled staff. Monahan told host Kain Warwick on the Nov. 18 episode of Unchained’s Uneasy Money that teams must crank visibility through the roof. She expects the next six months to be wild because AI can level up a team’s capabilities. A team defines a rule that prevents more than $200 million from leaving a contract under any circumstances to ensure an attack cannot drain the entire pool. If an attack reacts to autonomously, the team stays fine, and a breaker limits damage so a team is wrecked 5% of the way instead of 100% of the way.
| Security Layer | Coverage or Function |
|---|---|
| Audit | 90% coverage |
| Formal Verification | 5% coverage |
| Circuit Breaker | Automatic shutdown for the remaining 5% risk |
Recent exploits and AI threats
Recent attacks show why audits cannot catch everything. On Wednesday, K3 Capital paused the contracts behind sBOLD, its yield vault built on the Liquity V2 lending protocol, after security firm Blockaid said attackers drained about $3.43 million as of its first alert. On Monday, the perpetual futures exchange Lighter froze withdrawals after one trader’s leaked application programming interface key enabled a drain of funds. Lighter later said its Guardian safeguard caught the anomaly, and its Security Council returned about $5 million in USDC, a dollar-pegged stablecoin, to the trader before withdrawals reopened. Monahan notes North Korean hackers currently use AI to adjust social engineering scripts. She expects younger hackers to build entire campaigns around AI from scratch, which will change how these attackers optimize their technology. Monahan also noted in September that OpenAI could not monitor its agents.
