Sign InOpen Brain
AI EngineerVideoSource Linked

Teaching AI to Find Real Vulnerabilities — David Brumley, Bugcrowd

Cybersecurity agents need deterministic exploit oracles and tasks that reward finding distinct vulnerabilities, not self-reported wins or one crash. A capability ladder makes partial progress measurable.

AI Engineer · Aug 1, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Brumley describes reproducible vulnerable containers exposed through MCP, with setup, sandboxed file tools, and deterministic grading. He argues that crash-only checks miss exploit quality and duplicate findings; **50%** of DARPA Cyber Grand Challenge tasks reportedly contained unintended vulnerabilities, while AIxCC produced **18 unintended bugs**.

Practical Implication

For security-agent evals, grade concrete effects rather than model claims, deduplicate proofs by the vulnerabilities they trigger, and measure precision and recall across an audit set. A **16-capability ladder** can show where an agent stalls between triggering a flaw and building an out-of-sandbox exploit.

Agent-Ready Context
Brumley describes reproducible vulnerable containers exposed through MCP, with setup, sandboxed file tools, and deterministic grading. He argues that crash-only checks miss exploit quality and duplicate findings; **50%** of DARPA Cyber Grand Challenge tasks reportedly contained unintended vulnerabilities, while AIxCC produced **18 unintended bugs**.

For security-agent evals, grade concrete effects rather than model claims, deduplicate proofs by the vulnerabilities they trigger, and measure precision and recall across an audit set. A **16-capability ladder** can show where an agent stalls between triggering a flaw and building an out-of-sandbox exploit.

Even deterministic oracles encode a chosen definition of progress, and newly discovered bugs can change the ground truth. Publishing transcripts is also unresolved when a benchmark agent produces weaponized, previously private exploits.
Connected Context · Feed7 Judgment

This sharpens general agent-eval guidance for offensive security: success should be graded by reproducible effects, duplicate proofs should collapse to the vulnerability they exercise, and capability stages should separate triggering a flaw from achieving a real exploit. It also exposes a domain-specific limit of fixed ground truth: unintended vulnerabilities can invalidate labels, while transcript publication may disclose weaponized results.

Rethinking Environments for Long-Horizon Work — Rayan Garg, Theta SoftwareBoth reject proxy measures in favor of inspecting consequential state; here that principle becomes grading concrete exploit effects rather than model claims or nominal task duration.Verifiable Environments for AI in Biology — Kenny Workman, LatchBioThe biology candidate’s brittle-grader problem parallels security benchmarks where valid unintended vulnerabilities can defeat a fixed oracle, supporting human review when ground truth changes.Vending-Bench: Long-Horizon Agent Evals — Lukas Petersson, Andon LabsVending-Bench’s warning about agents exploiting evaluations is reinforced by security tasks containing unintended attack paths, making real-world checks and trajectory inspection important complements to deterministic containers.Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan EvaluationTyped-state gating and the 16-capability ladder share a structural remedy: withhold credit until required intermediate states are demonstrated, rather than rewarding an incomplete plan or crash-only result.
Context Map
benchmarksecurity#agent-evals#benchmark-integrity#agent-reliability
Uncertainty
Even deterministic oracles encode a chosen definition of progress, and newly discovered bugs can change the ground truth. Publishing transcripts is also unresolved when a benchmark agent produces weaponized, previously private exploits.