DARPA AI Cyber Challenge (AIxCC) Semifinals

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

Readiness band and full integer triple

AI-assisted assembly · derived results

Readiness band

Emerging

Primary summary from verified readiness

Evidence strength Weak

Decision

Agent-assisted assessment · independently reviewed · 2026-10-05

Consider the 2024 semifinal systems for supervised research pilots on selected C and Java projects.

Why: They produced useful vulnerability discoveries and patches under competition conditions, including a corroborated SQLite defect. Coverage was limited, and independent evidence does not establish reliable autonomous patching in production.
Next: Reproduce the original build on representative repositories and require functional regression tests and maintainer review before applying generated patches.

Constraints

Most seeded vulnerabilities were missed in the semifinal evaluation.
Generated patches require independent functional and regression validation.
Repository preparation and competition harnesses limit transfer to arbitrary codebases.
Production reliability of the original semifinal builds remains unproven.
Improvements in later finalist systems do not establish the capabilities of the 2024 builds.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Revision 4