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 specializes benchmark-integrity guidance for security agents: verify concrete exploit effects, prevent duplicate findings from inflating scores, and use a capability ladder to expose partial progress. It strengthens final-state grading but also limits confidence in deterministic oracles, because their definitions and ground truth can change, while useful trace publication may create security risk.

When Will The Benchmaxxing Plague End? — Nick Heiner, Surge AIThe deterministic exploit oracle is a security-specific response to weak verifiers and reward hacking, while the mutable ground truth confirms that even aligned grading needs scrutiny.Rethinking Environments for Long-Horizon Work — Rayan Garg, Theta SoftwareGrading triggered vulnerabilities and exploit effects implements Theta’s final-state emphasis, while the 16-capability ladder adds visibility into intermediate progress.Verifiable Environments for AI in Biology — Kenny Workman, LatchBioBoth show that domain-grounded evaluators must verify real task outcomes yet can misclassify valid behavior when the encoded oracle is too narrow.Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan EvaluationTyped-state gating and vulnerability deduplication address the same evaluator failure mode: agents should not earn higher scores through structurally invalid shortcuts rather than completing the intended task.
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.