Cyber Risks · Trail of Bits Blog

Trail of Bits

Summary: At DEF CON 33 in Las Vegas on August 8, 2025, DARPA announced Trail of Bits took second place and $3M in the AI Cyber Challenge finals with its Buttercup cyber reasoning system, behind Team Atlanta's $4M win and ahead of Theori's $1.5M. In the final round, which spanned 48 challenges across 23 open-source repositories and barred human interaction, Buttercup found 28 vulnerabilities and applied 19 patches, submitting proofs of vulnerability across 20 of Mitre's Top 25 Most Dangerous CWEs using only non-reasoning LLMs at a cost of $181 per point.

Key Takeaway: A fully automated system that finds and patches real vulnerabilities at $181 per point, built on cheaper non-reasoning models and downloadable to run on a laptop, moves automated vulnerability discovery from research demo to something your own engineering teams can run against their code.

Read the original at Trail of Bits Blog ↗

Topics
AI Security
Related