GitHub Overhauls Bug Bounty Program, Slashes Public Payouts Amidst AI-Driven Security Landscape Shift

Beginning July 27, 2026, GitHub will implement significant changes to its public bug bounty program, dramatically reducing payout amounts across all severity levels. This strategic pivot, announced by the software development platform, aims to curb the influx of lower-quality submissions and prioritize more impactful vulnerability disclosures. The move comes at a time of rapid advancement in artificial intelligence, which is increasingly being leveraged for both finding and potentially fixing software vulnerabilities, creating a complex new dynamic for bug bounty programs worldwide.

Under the revised public program, critical bug bounties will see their maximum payout reduced from $20,000-$30,000+ to a fixed $10,000. Similarly, high-severity findings will drop from a potential $10,000-$20,000+ to $5,000. Medium-severity reports will be capped at $2,500, down from $5,000-$10,000+, and low-severity findings will be reduced to $500 from $1,000-$2,500+. These new fixed rates represent a substantial reduction, with The Hacker News calculating that medium, high, and critical findings will be cut by at least 50%, and low-severity reports by approximately 59% when compared to the lower end of GitHub’s previous reward ranges.

Reports submitted and validated before the July 27th deadline, including those already in GitHub’s extensive triage queue, will be honored under the existing payout structure. This ensures fairness for researchers who have already invested time and effort into finding vulnerabilities under the prior terms.

GitHub stated that the restructuring is intended to "reduce noise" and foster a more efficient security ecosystem. The company articulated its new philosophy with the statement, "You don’t earn more by submitting more. You earn more by submitting better." This suggests a move towards valuing the depth and quality of research over the sheer volume of submissions. The goal is to provide established, high-caliber researchers with faster response times, more substantial rewards for exceptional work, and closer collaboration opportunities with GitHub’s dedicated security engineering team.

A Tale of Two Tiers: Public vs. VIP Programs

While the public program undergoes a significant payout reduction, GitHub is also formalizing its invite-only VIP tier. This exclusive program, reserved for its most trusted and high-performing security researchers, will offer substantially higher rewards. Under the VIP schedule, payments will be set at $1,000 for low-severity findings, $7,500 for medium, $20,000 for high, and $30,000 or more for critical vulnerabilities. This tiered approach clearly delineates the platform’s strategy: a broad public avenue for general vulnerability reporting and a more exclusive, high-reward channel for top-tier security talent.

Researchers can qualify for the private, invite-only VIP program by demonstrating consistent high-quality contributions. The current criteria stipulate reporting at least one critical, two high, four medium, or seven low-severity vulnerabilities. The announcement did not specify a precise time window for meeting these thresholds or guarantee an invitation upon qualification, indicating that further details regarding selection criteria will be published on GitHub’s HackerOne program page. GitHub has also indicated that it will enforce a HackerOne Signal threshold, though the specific level remains undisclosed. Researchers falling below this threshold will initially be limited to a maximum of four submissions. This aligns with HackerOne’s general policy for new researchers, which typically allows four trial reports per program within a rolling 30-day period.

The AI Catalyst: A New Era of Vulnerability Discovery

The timing of GitHub’s program overhaul is particularly significant, coinciding with a rapid acceleration in the capabilities of artificial intelligence in cybersecurity. AI is democratizing the discovery of potential vulnerabilities, making it cheaper and faster to generate candidate findings and automate code review processes. This surge in AI-assisted research has led to an exponential increase in the volume of submitted reports, often overwhelming security teams and bogging down triage processes with what GitHub terms "noise."

Just a day prior to GitHub’s announcement, Google unveiled Gemini 3.5 Flash Cyber, a specialized AI model meticulously fine-tuned for the discovery, validation, and patching of software vulnerabilities. Initially available to governments and trusted partners through Google’s CodeMender agent in a limited pilot program, this model represents a significant stride in AI-driven security. Google highlighted its efficiency, noting that the model can be invoked repeatedly to examine a vast number of code paths without the resource-intensive demands of larger frontier models. This makes it ideal for frequent repository scans, time-sensitive release reviews, and continuous commit-scanning pipelines.

Google’s internal testing demonstrated Gemini 3.5 Flash Cyber’s superior performance, identifying 55 unique confirmed V8 engine issues, surpassing mainline Gemini 3.5 Flash (47 issues) and Claude Opus 4.6 (36 issues). Furthermore, Google reported that its Cloud Vulnerability Research team leveraged the model to uncover remote code execution (RCE) flaws in public APIs and a memory-corruption vulnerability in a critical production service within a mere two hours. The model even generated a "100%-reliable" RCE exploit that successfully bypassed Address Space Layout Randomization (ASLR) and Write XOR Execute (W^X) protections. While these figures are compelling, they are self-reported by Google and await independent verification.

The ability of internal security teams to provide AI agents with detailed context, project-specific threat models, and tailored validation environments is transforming the security landscape. Systems like OpenAI’s Codex Security can now test findings, generate functional proof-of-concept exploits, and even propose fixes that align with system intent and surrounding behaviors. This allows for continuous security testing integrated directly into development workflows, rather than relying solely on scheduled assessments or external bug bounty submissions. While AI may not entirely replace human penetration testers, it is significantly streamlining source-code review, test generation, and initial vulnerability validation. Human expertise remains crucial for complex tasks such as chaining vulnerabilities across trust boundaries, identifying business logic flaws, modeling realistic attack scenarios, and proving material impact.

Precedent and Parallel Trends

GitHub’s move is not an isolated incident. The challenges posed by AI-generated noise have already prompted other significant shifts in the bug bounty community. Notably, Daniel Stenberg, the maintainer of the widely used curl project, ended its cash bug bounty program in January 2026. This decision was attributed to a decline in the confirmed vulnerability rate below 5%, a consequence of an overwhelming influx of AI-generated reports. While curl later returned to HackerOne, Stenberg observed in April 2026 that submission rates had doubled, and approximately 15-16% of reports were confirmed vulnerabilities. He noted that nearly every submission appeared AI-assisted, but a significant portion were now of high quality, indicating a potential evolution in AI’s utility.

The collective trends – GitHub’s program restructuring, Google’s advanced AI models, and curl‘s experience – highlight a fundamental shift: AI can both inundate maintainers with low-value reports and empower capable researchers to operate with greater efficiency. Simultaneously, internal teams can leverage AI to scrutinize codebases more rigorously and frequently.

The Value Proposition: Discovery vs. Impact

In this evolving landscape, the ease of generating plausible-looking candidate findings has increased dramatically. However, the critical constraints remain in the areas of triage, exploit verification, understanding product context, responsible disclosure, and remediation. Truly scarce are reliable, sophisticated exploits, attack chains tailored to specific product architectures, or vulnerabilities that exploit subtle misunderstandings of a vendor’s security model.

GitHub’s revised strategy, with its reduced public rewards and increased reliance on signal thresholds and invite-only tiers, appears designed to address this scarcity. While these measures may suppress automated noise and discourage low-effort submissions, they also present potential barriers for emerging security researchers who lack an established track record on platforms like HackerOne. For a new researcher on HackerOne, the limited four-report program allowance leaves little room for error, unfamiliarity with a platform’s specific security posture, or the submission of a legitimate finding that might be initially misjudged or scored lower than its actual impact.

The emphasis on an invite-only structure for the VIP tier also concentrates GitHub’s closest researcher relationships among those who have already proven their mettle within the program. This approach is likely to enhance response speed and report quality by fostering deeper collaboration. However, it may also inadvertently narrow the diversity of researchers examining the platform, potentially diminishing one of the key advantages of a broad public bug bounty program – a wider array of perspectives and approaches to security testing.

This latest policy adjustment follows a broader strategic shift initiated by GitHub in May 2026. That policy change mandated the submission of working proof-of-concept exploits, demonstrated impact, pre-submission validation, and a more stringent adherence to GitHub’s defined scope and ineligible findings. GitHub has publicly stated its embrace of AI-assisted security research, noting its own internal use of AI across various security programs. However, the platform emphasizes that researchers remain ultimately responsible for verifying the accuracy and validity of any findings generated by their tools. As GitHub succinctly put it, "The tools don’t matter. The quality of the work does."

As of July 22, 2026, a review by The Hacker News indicated that GitHub’s public rewards page still reflected the older payout structure, listing $20,000-$30,000+ for critical reports. Similarly, the FAQ section retained previous VIP eligibility criteria, requiring at least $20,000 in earnings and two submitted reports within the preceding two years, with the caveat that meeting these criteria did not guarantee an invitation, and candidate reviews occurred quarterly.

GitHub’s new pricing structure starkly illustrates the shift in its security bounty strategy: a public critical finding now warrants $10,000, while a critical finding within the VIP program can yield $30,000 or more. The era of easily discovering potential vulnerabilities is becoming more commoditized. The true premium, it appears, now lies in delivering verified, product-specific impact and sophisticated security analysis that AI alone cannot yet fully replicate. This evolving landscape underscores the increasing importance of human ingenuity and deep technical understanding in the ongoing battle for software security.