Four Bug Bounty Programs Cite Rising Submissions: GitHub Tiered Pay, Apple Capped Reports, Google Paused a Class, Microsoft Paid Over $20 Million
GitHub, Apple, Google and Microsoft each describe rising bug bounty submissions but report different steps; the sources word the cause differently, and none of the cited sources gives the share produced with AI.
Editor's Note ·
- Clarification:
- The article says the quote from Google that The Hacker News reports ("a significant rise in automated submissions, the vast majority of which are not valid") "may come from the notice or a follow-up post". The editor read the notice as it appears in Google's public GitHub copy of the OSS VRP rules (google/bughunters, commit f8bf23a, September 30): it announces that Google is no longer accepting product vulnerabilities from October 1, 2026 and commits to a Q1 2027 update, and it gives no reason and does not contain the quote. The October 1 X post does not contain it either, so the wording "automated" rests on The Hacker News's report alone.
- Clarification:
- The Implications section says "GitHub gives newcomers up to four initial submissions". GitHub's July 22 post says "HackerOne's platform gives researchers who don't meet the threshold up to four initial submissions".
- Clarification:
- The Implications section places Google's move from a proof requirement at "March 20". That is the date of InfoWorld's report; the cited sources do not give the date of Google's own announcement, and The Hacker News says only that Google began requiring stronger proof for some tiers in March 2026.
Background
GitHub, Apple, Google and Microsoft each describe a rise in bug bounty submissions, and the steps attached to those statements differ. The accounts also differ in who speaks: GitHub and Microsoft published their own posts, while outlets relay Apple’s and Google’s statements. This analysis compares the four cases and labels which statements are the operators’ own, which are an outlet’s, and which are the Herald’s reading.
Two of the programs had changed their rules earlier in 2026. On March 20, InfoWorld reported that the team running Google’s Open Source Software Vulnerability Reward Program (OSS VRP) was “increasingly concerned about the low quality of some AI-generated bug submissions”. It quoted Google as saying it would “now require higher-quality proof (like OSS-Fuzz reproduction or a merged patch) for certain tiers”. InfoWorld’s opening sentence says Google “will no longer accept AI-generated submissions”; the quoted Google statement describes a proof requirement for certain tiers, and the Herald relies on the quoted text.
On May 15, GitHub published a post on its bug bounty standards that described “a sharp increase in submissions that don’t demonstrate real security impact” and also said “We welcome AI in security research”. It asked researchers to validate the output of scanners, static analysis and AI assistants before submitting, and wrote: “The tools don’t matter. The quality of the work does.”
Current situation
GitHub: tiers and a signal requirement
On July 22, GitHub announced a restructured program for reports filed on or after July 27, 2026. The public table pays $250 for low, $2,000 for medium, $5,000 for high and $10,000 for critical findings, with static amounts instead of ranges. A permanent invite-only VIP program pays $1,000, $7,500, $20,000 and $30,000+ for the same severities. GitHub says it is implementing a HackerOne signal requirement “To reduce the volume of low-effort and AI-generated reports”, under which researchers below the threshold get up to four initial submissions. The Register covered the change on July 23 and describes GitHub as Microsoft-owned; its account of the motive restates GitHub’s statement. The Herald previously reported the change; this analysis does not repeat that article’s comparison with the earlier pay ranges, on which a correction was filed.
Apple: a cap on submissions
MacRumors wrote on August 4, “according to The Financial Times”, that Apple limited the number of vulnerabilities researchers can submit, and says Apple described a high volume of poor-quality submissions from amateur bug hunters using AI. It describes a cap on the number of open submissions a researcher can have, and says researchers can request an increase. The Decoder, writing on August 2, also cites the Financial Times and adds a 30-day cooldown period that MacRumors does not mention. Both outlets restate the same Financial Times report, which the Herald could not read, so Apple is one underlying account carried by two outlets.
Both outlets describe one consequence: the cybersecurity startup Bynario, which they say used ChatGPT to find a macOS vulnerability, could not report it because of the limit. MacRumors adds that Bynario found more than 50 macOS bugs in three weeks, that Apple has since been in contact with the startup and is reviewing its submissions, and that Apple “too has turned to AI for parsing submissions”.
Microsoft: a record year and a wider scope
Microsoft’s year-in-review post, dated August 3, reports more than $20 million awarded to 562 researchers, against $17 million to 344 researchers the year before. It says the program saw “a notable increase in submission volume during the second half of the year, reflecting both strong engagement from the research community and the growing use of AI to support security research”. The Register gives the program year as July 1, 2025 to June 30, 2026; if that is the “year” in Microsoft’s sentence, the second half would be January to June 2026, a reading that is the Herald’s. The Register also says Microsoft widened its program in December 2025 under what it calls “In Scope By Default”, and writes that “You could argue that the numbers do not represent a fair fight”. That analysis of the scope change is the outlet’s own, while its account of the AI attribution restates Microsoft’s post. The Herald reported the program’s results earlier.
Google: one class of report paused
The Hacker News reported on October 6 that, since October 1, the OSS VRP no longer accepts product vulnerability reports for rewards. Reports about supply chain compromises are still accepted, as are reports filed before October 1, and the rules notice commits Google to an update in the first quarter of 2027. The outlet names Go, Angular and Protocol Buffers among the affected projects. It quotes Google as citing “a significant rise in automated submissions, the vast majority of which are not valid”, and notes that Google gave no figures and did not say whether the submissions were produced with AI tools.
Tom’s Hardware wrote on October 3 that Google suspended submissions “over an influx of invalid AI-driven reports”, and says engineers and maintainers were “reportedly” overwhelmed by thousands of poorly written reports. The two outlets therefore differ on the cause: The Hacker News says AI is not stated in Google’s wording, while Tom’s Hardware writes AI. InfoWorld’s March report above shows the program’s team had already voiced concern about AI-generated submissions, so Tom’s Hardware’s wording has a basis, but it is the outlet’s characterisation of the October pause. The Hacker News attributes the quote to Google’s October 1 post on X; the text of that post the Herald read does not contain it and gives no reason, and the Herald could not read the rules page, so the quote may come from the notice or a follow-up post. The Herald earlier reported the pause and placed it beside project disclosure rules in an analysis.
Tom’s Hardware also writes that Intel suspended a bug bounty program that paid up to $100,000 per flaw, and that the company “did not officially confirm AI-generated reports as the reason for the move, but experts suspect this is the case”. That case rests on one outlet and is not counted among the four.
Implications
The points below are the Herald’s reading of the sources, not a forecast.
The steps differ. GitHub changed the payout table and who can submit freely, Apple limited how many submissions a researcher can hold, Google paused one class of report, and Microsoft reported a record payout under a wider scope. The cited sources show a two-step sequence for two programs: Google went from a proof requirement (March 20) to a pause (October 1), and GitHub from validation standards (May 15) to tiers and a signal requirement (announced July 22, effective July 27). The cited sources describe no earlier step for Apple.
The programs word the cause differently. By the Herald’s count of the wording each source attributes to the operator, GitHub (“AI-generated”), Apple (amateur bug hunters “using AI”, per MacRumors) and Microsoft (“the growing use of AI”) name AI in three of the four cases; Google’s wording as quoted is “automated”. Microsoft presents AI use as one reason for more submissions alongside “strong engagement”, while the other three describe low-effort, poor-quality or invalid reports. The Herald also notes that GitHub, which The Register calls Microsoft-owned, restricted newcomers while Microsoft’s program reported record payouts and its most researchers, within two weeks of each other (GitHub announced on July 22, Microsoft’s post is dated August 3); no cited source says the programs coordinated or compared notes.
No cited source measures the share. None gives a program’s total submission count or the share generated with AI. Microsoft’s post gives counts for subsets only: nearly 700 reports at Zero Day Quest and more than 300 additional reports from the wider scope. Tom’s Hardware reports “thousands” of poor reports without naming a source and says “reportedly”; The Hacker News says Google’s post gave no figures. Whether the new limits reduced noise is not reported by any cited source.
Exceptions are part of the designs, with costs. GitHub gives newcomers up to four initial submissions and Apple lets researchers request an increase, yet the Bynario case, as MacRumors tells it, shows a limit blocking a report of a privilege escalation exploit that could give an attacker unrestricted access to a Mac. For developers whose projects sit inside these programs, Google’s pause means product vulnerabilities in projects such as Go and Angular no longer earn an OSS VRP reward, and The Hacker News notes that Angular’s security policy names no other reporting channel.
What the sources do not establish
- Apple’s exact limit, the 30-day cooldown (reported by The Decoder only) and the date either took effect are not given by the cited outlets, and the Financial Times report could not be read.
- Google’s reason in its own words could not be read first-hand; the AI attribution rests on Tom’s Hardware.
- No cited source reports how many submissions to any of the four programs were AI-assisted, or what happened to report quality after the changes.