a5fa51fe-7
Taylor Monahan: Audits Cover 90% of Risk — Teams Need Circuit Breakers
Security researcher Taylor Monahan says audits cover 90% of risk and formal verification another 5%, urging teams to add circuit breakers and monitoring as AI-driven probing accelerates.

Outputs
Taylor Monahan estimates audits cover ~90% of risk, formal verification another 5%, leaving 5% that can cause total failure
She recommended circuit breakers, hypothetically capping outflows at $200 million per contract, on the Nov. 18 Uneasy Money episode
K3 Capital paused sBOLD (Liquity V2) contracts after ~$3.43 million was drained during the week of Nov. 16
Lighter froze withdrawals after a leaked API key incident; its Guardian safeguard detected it and ~$5 million in USDC was returned
Monahan said North Korea-linked hackers already use AI for tasks like social engineering scripts
Security researcher Taylor Monahan estimates that a smart contract audit covers roughly 90% of a system's risk and formal verification adds another 5% — leaving a residual 5% that can still produce a complete failure. Her prescription, delivered on the Nov. 18 episode of the podcast Uneasy Money: crypto teams must build the monitoring and automatic safeguards that catch what audits cannot.
Monahan said her guidance has shifted toward monitoring over roughly the past six months. She urged teams to increase visibility into system activity and to respond autonomously when an attack is detected, rather than relying on human intervention while funds drain.
The concrete mechanism she recommended is a circuit breaker: a rule that pulls the plug when a defined condition is met. Her hypothetical example is preventing more than $200 million from leaving a single contract. In practice, such a cap would have limited most of the largest exploits in recent memory to a fraction of their actual losses.
What does this change for practitioners?
The message reframes where engineering budgets should go. Instead of treating an audit report as the end of security work, teams should assume a compromise scenario and build the tripwires for it. Monahan's framing effectively splits the problem: audits and formal verification address what reviewers can reason about; monitoring and breakers address what they cannot.
Artificial intelligence cuts both ways in that equation. Monahan said AI can help smaller teams, which lack dedicated security staff, build systems that watch for anomalies and react to attacks. At the same time, AI agents allow attackers to probe projects far more frequently than manual review permitted.
She added that hackers tied to North Korea already use AI for small tasks, such as adjusting social engineering scripts. Attackers have not yet optimized their use of AI, she said, but she expects younger North Korean operators to build entire campaigns around it.
The week's incidents made the argument
Monahan's comments landed during the week of Nov. 16, amid two live security incidents that illustrate both failure and success modes.
On Wednesday, K3 Capital said it had paused the contracts behind sBOLD, a yield vault built on Liquity V2. About $3.43 million had already been drained by the time the first alert fired — the detection-then-response gap Monahan wants teams to close.
On Monday, the exchange Lighter froze withdrawals after an attacker took funds using a leaked API key belonging to one trader. Lighter said its Guardian safeguard detected the anomaly and its Security Council returned roughly $5 million in USDC, a dollar-pegged stablecoin, to the affected trader. Withdrawals have since reopened.
The contrast is the point. Lighter's automated Guardian layer caught an incident in progress and contained it; the sBOLD vault lost $3.43 million before anyone reacted.
For teams shipping contracts in the current environment, the operating assumption is that probing is continuous and increasingly automated. Circuit breakers with hard dollar limits, anomaly monitors and pre-authorized response councils are shifting from best practice to baseline expectation — and the gap between projects that have them and projects that do not is now measurable in millions per incident.
via tokenpost.com (Source)
More from Nathan Brooks
Show full bio
Market editor covering money, funding and market shifts around crypto and web3 industry at Mempool Brief.
1,651 articles