AI Is Flooding the Bug Bounty Market and Pushing Prices Down
Security researchers who make a living finding software flaws are earning less per discovery as AI tools drive a wave of reports that is reshaping the entire bug-hunting industry.
Key points
- HackerOne CEO Kara Sprague says bug report volume roughly doubled year over year at her platform.
- TrendAI's Zero Day Initiative recorded a 450% spike in submissions in April 2025 before volumes eased back.
- Bugcrowd saw a 300%-plus surge during a three-week window, settling at roughly double historical norms.
- A macOS full TCC bypass that once paid roughly $30.5K may now be worth around $5K, according to one veteran hunter.
- Bugcrowd CEO Dave Gerry says 82% of researchers now use AI to help with their work.
There's a corner of the security world where independent researchers quietly hunt for flaws in popular software. When they find something, they report it to the company responsible. The company pays them a reward, called a bug bounty, and fixes the problem before criminals can exploit it. For many, that income is their living.
That living just got harder.
What changed?
AI tools, particularly large language models (the same category of software behind ChatGPT), are now helping researchers find flaws faster and in far greater numbers. The result is a flood of reports hitting the platforms that manage bug bounty programmes.
Submission volumes have roughly doubled at HackerOne, according to CEO Kara Sprague. At TrendAI's Zero Day Initiative (ZDI), a programme that buys and discloses security flaws on behalf of vendors, the rate jumped 450% year-over-year in April 2025 before moderating. Bugcrowd saw a 300%-plus surge in a single three-week period, though volumes have since settled at roughly double historical norms. We first reported on HackerOne's economics on 3 July 2026, when Microsoft's $20 million payout year raised questions about whether platforms were handling researcher work fairly.
More supply means lower prices. That's basic economics, and it's hitting researchers hard.
How much are prices actually falling?
The drop is real. Bug researcher Wojciech Reguła, who focuses on Apple's macOS, told Dark Reading that bounty amounts have clearly declined in his area. A full TCC bypass, meaning a flaw that lets an attacker override Apple's privacy controls over which apps can reach your photos, microphone or location, used to pay around $30.5K. The same class of flaw now earns roughly $5K. A more limited bypass, such as silently accessing a user's photos, dropped from around $5K to $1K.
| Flaw type | Previous payout | Approximate payout now |
|---|---|---|
| Full TCC/privacy bypass (macOS) | ~$30.5K | ~$5K |
| Limited TCC bypass (e.g. Photo access) | ~$5K | ~$1K |
Dustin Childs, head of threat awareness at ZDI, was direct: "The $2,000 to $50,000 bugs, I think those are going to become very scarce, or the price is going to be pressed down."
Is the whole industry collapsing?
Not quite. Sprague told Dark Reading that bounty payments at HackerOne rose 25% in the first half of 2025 compared with the same period a year earlier, and the number of researchers clearing $100,000 is up 25% too. The market is growing overall even as individual findings shrink in value.
What's changing is the shape of the work. Bugcrowd's Gerry expects bug hunting to become a volume game, where researchers use AI to find more flaws at lower value rather than fewer at higher value.
Independent researcher Ashish Kunwar describes using AI as a co-pilot, not an autopilot. He built an internal tool pairing static code analysis with a local AI model to speed up review. But he draws a firm line. "The judgment calls," he told Dark Reading, "'is this real, is it exploitable, does it cross a boundary, how do I prove impact,' those are mine."
Should you worry about quality?
The bigger short-term problem is noise. When AI generates security reports automatically, many are wrong or irrelevant. In January 2026, curl creator Daniel Stenberg shut down the bug bounty programme for curl (software that transfers data across the internet) after seven years. Historically more than 15% of submissions confirmed real flaws; by 2025 that figure had fallen below 5%, choked by what Stenberg called "AI slop reports." Apple has begun pausing reporting privileges for users who repeatedly send ineligible submissions. Our 14 August story on the US government's flaw database found the same dynamic at scale: a 72% surge in reported flaws straining the humans who have to review them.
For ordinary readers, none of this creates new risks today. A slower, noisier pipeline does mean some genuine flaws may take longer to reach a fix, so keeping software updated remains the single most effective precaution available.
