GitHub cuts bug bounty payouts as AI reports flood in

GitHub's restructured bug bounty program takes effect today, and the shape of it is more instructive than the numbers. Public payouts get cut across every severity tier, top rewards move behind an invitation-only door, and new researchers face submission limits until they've built a track record. The driver is a flood of AI-generated vulnerability reports that look plausible, cost nothing to produce, and take a human to disprove. If you run any intake queue at all, that's your problem too.

What actually happened

Per GitHub's own post, the public program now pays flat amounts: $250 for low, $2,000 for medium, $5,000 for high, and $10,000 for critical. The high tier previously topped out at $12,500. The new invitation-only VIP tier pays $1,000 / $7,500 / $20,000 / $30,000-plus across the same severities, along with faster responses and direct access to GitHub's security team.

Getting into VIP requires a demonstrated track record — one critical finding, or two highs, four mediums, or seven lows. New researchers without history get a small number of initial submissions before hitting limits, a HackerOne signal gate aimed at automated and low-effort volume. Reports filed before today are grandfathered under the old structure.

GitHub frames the change around a growing queue and the noise inside it. The Register's coverage puts it plainly: the security team is being buried by AI-generated submissions.

Why AI-generated report volume matters for your business

Read what GitHub actually did. They didn't build a better filter for bad reports. They repriced the queue and gated it on reputation — because when generating a submission costs nothing, no content filter wins on volume. Verification is the scarce resource, so they moved payment toward the people whose submissions are cheap to verify.

That pattern is landing on every intake you own. Job applications. Contact forms. Support tickets. Vendor pitches. Warranty and return claims. Each one assumed a human effort cost on the sender's side that no longer exists.

Three things worth doing before it hits you. Instrument the queue before you touch the rules — measure median time-to-disposition and what share of items resolve as noise, because you can't tell overload from a bad week without a baseline. Move gating from content to reputation: a known customer, a verified account, a past-transaction record. Those are cheap signals a generator can't fake and they don't get fooled by better prose. And route by verification cost, not by claimed severity — an item that takes two minutes to confirm should never sit behind one that takes an hour, no matter how alarming its subject line reads.

The submissions aren't going to get worse. They're going to get better-written, which is the actual problem.

Sources: GitHub Blog, The Register.