Starting July 27, 2026, GitHub will cut community bug bounty payments by at least half at all severity levels. Key finds will drop from $20,000-$30,000+ to a fixed $10,000, while its invitation-only VIP category will pay $30,000 or more.
Reports submitted prior to that date, including those already in GitHub’s growing third row, will retain the previous payment terms.
GitHub said the changes are intended to reduce noise while giving senior researchers faster answers, higher rewards, and closer access to their security engineering team.
“You don’t get more money for sending more,” the company said. “You earn more by posting better.”
The public plan ranges from flexible scope to fixed fees:
- Low: $250, down from $617-$2,000
- Average: $2,000, down from $4,000-$10,000
- High: $5,000, low from $10,000-$20,000
- Value: $10,000, down from $20,000-$30,000+
Hacker News calculated that the new public values are 50% lower for medium, high, and critical finds and about 59% lower for low-severity reports when measured against GitHub’s previous range.
GitHub said fixed payments should remove uncertainty and triage overhead, although it may offer discretionary bonuses for special work.
The VIP program sets payouts at $1,000 for low-risk findings, $7,500 for medium, $20,000 for high, and $30,000 or more for critical risk.
Researchers can find a classified program by reporting at least one serious, two high, four moderate, or seven low severity risks. The announcement does not specify a time window for meeting those limits or say whether eligibility warrants an invitation. GitHub said the full terms will appear on its HackerOne community page.
GitHub did not disclose the limit HackerOne Signal will use. The company says its affiliate researchers will receive up to four initial presentations.
Separately, HackerOne’s standard rules give new researchers four trial reports per program within a 30-day window.
When Your Own AI First Finds a Bug
GitHub’s report controls come as AI makes candidate findings cheaper to produce and code reviews cheaper to replicate. Additional researchers can reveal potential findings, while internal teams can scan code, confirm issues, and fix feeds in releases and pipelines before an external report arrives.
A day before GitHub’s announcement, Google launched Gemini 3.5 Flash Cyber, a lightweight model optimized for detecting, approving, and patching software vulnerabilities. Google said the model would initially be available only to governments and trusted partners through CodeMender, its code protection agent, as part of a limited pilot.
Google said the model can be asked repeatedly to test multiple code paths without using a large parameter model for every attempt. The company deploys it for frequent testing in the repository, time-sensitive launch reviews, and scanning commitment pipelines.
In tests conducted by Google, the Gemini 3.5 Flash Cyber received 55 confirmed V8 issues, compared to 47 for the original Gemini 3.5 Flash and 36 for the Claude Opus 4.6.
Google separately said its Cloud Vulnerability Research team used the model to detect remote code execution errors in public APIs and a memory corruption error in a critical production service within two hours. The model then produces what Google described as a 100% reliable RCE exploit that bypasses ASLR and W^X. Benchmark statistics and productivity exploits are reported by Google and have not been independently verified.
An internal security team can provide an agent repository context, a project-specific threat model, and an authentication environment designed for the operating system. Systems like OpenAI’s Codex Security can then analyze the findings, generate functional proofs of concept, and propose fixes that are consistent with the system’s intent and surrounding behavior.
Work can happen during development and on all relevant commitments, rather than waiting for a planned test or external report. AI doesn’t replace penetration testing, but source code review, test generation, and first-pass validation become easier to automate.
Human testers retain added value when the job requires binding weaknesses across trust boundaries, identifying business logic failures, modeling realistic attack methods, and proving material impact.
Curl maintainer Daniel Stenberg ended the project’s bug bounty at the end of January 2026 after the confirmed vulnerability rate dropped below 5% amid an increase in AI-generated spam reports.
In April, after curl had finished the prize money and returned to HackerOne, the reports were almost double the 2025 average and 15-16% was confirmed as a risk. Stenberg said almost all reports seem to favor AI and most are at a high level.
Combined, GitHub’s report controls, Google’s iterative call model, and curl’s incremental traffic volume point to the same switch. AI can flood maintainers with garbage, but it can also make skilled researchers faster and allow internal teams to test more code, more often.
Finding a candidate that seems to sound like more and more. Testing, exploiting evidence, product context, disclosure, and correction are often delayed. Credible exploits, product-specific attack chains, or cross-border discoveries by a misunderstood vendor remain rare.
Signal requirements and low community rewards may suppress automatic noise, but they can also make entry difficult for skilled researchers without an established HackerOne history. For HackerOne’s new researcher, the program’s four-report limit leaves little room for error, unfamiliarity with GitHub’s security model, or legitimate discoveries that fall short of expectations.
The invitation-only structure also fosters GitHub’s close researcher relationships among people who are already successful within the program. That may improve the speed and quality of reporting. It may also reduce the number of people checking out the platform, one of the main benefits of a public benefit system.
The reorganization follows a May 2026 policy change that required effective proof of concept, demonstrated impact, validation before deployment, and close attention to GitHub scope and inappropriate findings.
GitHub said it welcomes AI-assisted security research and already uses AI in all of its internal security programs. Researchers remain responsible for the reproducibility and validation of anything produced by their instruments.
“Tools don’t matter,” says GitHub. “The quality of the work does.”
As of July 22, a review by The Hacker News found that GitHub’s rewards page still lists $20,000-$30,000+ for critical reports, while its FAQ has retained previous VIP eligibility tests of at least $20,000 earned and two reports submitted within the past two years. The FAQ also stated that meeting those criteria does not guarantee an invitation and that GitHub will review candidates quarterly.
