Snyk's Eli Cohen: fastest AI attack took 4 minutes, security needs agent teams
AI Engineer · youtube · 2026-10-08
In an AI Engineer World's Fair talk, Snyk's Eli Cohen (Helios co-founder) argues security must go offensive: attackers use frontier models to chain vulnerabilities in as little as 4 minutes, while annual pen tests can't keep up. He contrasts SAST, DAST and pen testing, then outlines continuous offensive security — AI pen testing on every PR via an agent team (orchestrator, recon, vulnerability hunters, exploit-validating judge, remediation), plus four questions to vet AI pen-testing vendors.
More from coding & agent
- Developer builds agent-native Adobe clone, uses Opus 5.5 to reproduce original art — teortaxesTex · 2026-10-08
- User Reports Astra Agent Degradation: Grabbing Months-Old Checkpoints 10-15 Times — almmaasoglu · 2026-10-08
- Months of simulation code loses to a from-scratch rewrite with Opus 5.5 — IgorCarron · 2026-10-08
- "Software is over": AI developer takes aim at Adobe with open source clones — HarrySurden · 2026-10-08
- Pi Durable vs. OpenCode: How the Newest Coding Harness Breaks the Architecture Mold — HankYeomans · 2026-10-08
- Dev built an agentic pipeline where Hermes designs t-shirts and runs a Shopify store — alexcovo_eth · 2026-10-08