How Figma Stays Ahead of Vulnerabilities With Agents

Rohan Sharma, Liam Buchan and Dave Martin on running security agents at three points in Figma’s pipeline, all driven by one shared document. Agents review every PR, inject just-in-time guidance while code is being written when it touches risky patterns like new routes or permission policies, and audit the ten-year-old monorepo in budget-aware shards. The claim that ties it together is that the policy is the threat model: writing the trust boundaries, accepted risks and precedents down in one place was the work, and the three pipeline stages are just consumers of it.

That document is small. Ninety-nine lines, 2,560 words and 68 precedents. Its measured effect is large: without the policy the agents got 44.4% payout-weighted recall, with it 64.2%.

The ordering advice is precision before recall. Week one was bad, 4 valid findings out of 27, about 15%. Within a month, iterating the policy pushed precision to 80% on a two-week lookback, past their 70% bar. Recall is measured against a growing corpus of 66 real vulnerabilities that got past human review, drawn from their HackerOne programme and internal incidents, and current frontier models catch 75.8% of them. The first full repo audit turned up more than a hundred latent vulnerabilities, including two criticals that traditional SAST tools had missed.

They are explicit that this is not research-grade evaluation: we’re not chasing pass^k confidence intervals or building holdout sets. We need assurance that the controls work and telemetry that shows when performance slips. The shift they describe is security engineers moving from triaging one bug at a time to writing the policy that catches hundreds.