The Extended Brief

How we use /goal to find bugs in Patch the Planet

Brief by The AI News AI newsroom · Jul 31, 2026, 1:06 PM EDT edition

Original reporting by Trail of Bits · published Jul 28, 2026, 7:00 AM EDT

Updated Aug 1, 2026, 11:28 AM EDT

Trail of Bits demonstrates that letting Codex write its own goal prompts significantly improves autonomous bug hunting in critical open-source codebases.

Key points

• Codex identified every submitted Rust bug, including a soundness hole patched in version 1.98. source ↗
• The Patch the Planet initiative uses the /goal feature to audit major open-source codebases. source ↗
• The tool converted past CVEs into Semgrep rules and flagged eleven variant hits. source ↗
• It discovered two potential high-severity privilege-escalation vulnerabilities within the Keycloak SAML component. source ↗
• Engineers achieve better results by having Codex generate its own goal prompts from threat models. source ↗

Practical applications

• Try having Codex generate its own /goal prompts from a threat model of your codebase rather than hand-writing them, which Trail of Bits found produces better results.
• Convert your organization's past CVEs into Semgrep rules and use an agent to hunt for variant bugs, mirroring the approach that flagged eleven variant hits.
• Review your Keycloak SAML configuration and track the two potential high-severity privilege-escalation findings if you rely on that component.
• Calibrate scope and the number of requested outcomes per run, which Trail of Bits identifies as key variables for useful agentic bug hunting.

Context

Patch the Planet is a joint Trail of Bits and OpenAI initiative that points Codex, an AI coding agent, at major open-source codebases to find and fix bugs, using a /goal feature that directs long-running autonomous audits. Results so far include identifying every submitted Rust bug — among them a soundness hole patched in version 1.98 — eleven variant hits from CVEs converted into Semgrep rules, and two potential high-severity privilege-escalation findings in Keycloak's SAML component. The headline methodological finding is that letting Codex write its own goal prompts from threat models outperforms hand-crafted prompts.

What to watch

• Confirmation and severity assessment of the two potential Keycloak SAML privilege-escalation vulnerabilities.
• Whether Patch the Planet's disclosed-and-fixed bug count grows enough to validate agentic auditing across the open-source ecosystem.

Related briefs

• Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
• DeepSeek launching v4.1 flash cheaper and more capable than v4 pro
• AI coding startup Cognition raises $2B at $48B valuation as revenue nears $900M
• Latest open artifacts (#24): Motif-3, GLM-5.3, Hy4-preview and open model licenses

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.