AI 與軟體安全
Low-Quality AI Reports Drive Up Triage Costs as GitHub Bug Bounty Program Adopts Public and Invite-Only Tracks Today
Starting July 27, GitHub is using a HackerOne signal threshold to restrict submitters without an established track record, while moving higher rewards and faster responses to an invite-only VIP program. The changes directly address the rise in low-effort and AI-generated reports, but may also raise the barrier for new researchers submitting valid vulnerabilities.
GitHub’s revamped bug bounty program applies to reports submitted on or after July 27, 2026. The central change splits the previous single public program into two permanent tracks: public and VIP. GitHub says its backlog has continued to grow, creating a particular need to reduce noise from low-effort and AI-generated reports so its security team can focus on reproducible vulnerabilities with real-world impact.
The public program now uses fixed rewards of $250, $2,000, $5,000, and $10,000 for low-, medium-, high-, and critical-severity vulnerabilities, respectively. It also introduces a HackerOne signal threshold; researchers who have not yet met the threshold receive up to four initial submission opportunities to establish a track record. The VIP program is invite-only, with corresponding rewards of $1,000, $7,500, $20,000, and at least $30,000. It also promises faster responses and more direct collaboration with GitHub’s security engineering team. Researchers can qualify with one valid critical-severity finding, two high-severity findings, four medium-severity findings, or seven low-severity findings.
This is not an update to a model, scanner, or API, but it is an important technical signal as AI-assisted vulnerability research enters production workflows. Generative tools have reduced the cost of producing seemingly complete reports, without similarly reducing the effort required to verify exploitability, create a minimal reproducible example, and define the impact of an attack. When submissions can be automated at scale but triage still requires humans to analyze HTTP traces, permission boundaries, and affected assets, bug bounty platforms face an asymmetric drain on resources.
Teams using agents for security testing therefore should not send model output directly into vulnerability disclosure channels. A more reliable pipeline should require replayable requests, clearly stated prerequisite permissions, affected versions, the actual security boundary crossed, and human verification, while deduplicating repeated or purely theoretical findings. Key questions are whether the signal rules will genuinely improve triage times, and whether lower public rewards and a limited number of initial submissions will drive away new researchers who lack an established reputation but possess valid findings. GitHub has not disclosed the share of reports generated by AI, the false-positive rate, or the size of its backlog.