Ask anyone who runs a bug bounty program what changed this year, and you will hear the same answer: the inbox.
Raj Handa, Senior Cybersecurity Engineer in offensive security at Workday, leads Workday’s pentest program and runs its bug bounty. When I asked him on Scale To Zero what his mornings look like, he started with exactly this:
“These days a huge share of it is AI-generated slop, which is basically pages of hallucinated exploit narratives that really sound brilliant until you realize that there is zero reproducible bug.”
His first job of the day: “separating genuine vulnerabilities from confident noise.”
This post covers what Raj told me about keeping a program healthy when that noise keeps growing.
The curl Signal
Open source maintainers felt this first. In June, curl maintainer Daniel Stenberg announced the curl “summer of bliss”: no vulnerability reports accepted or handled during July 2026, with curl’s HackerOne submission form paused until August 3. He wrote that the project had been “under a huge pressure for the last four months or so.”
Then in October, SC Media reported that curl would require low and medium severity reports to be filed in its public GitHub repository starting October 26, keeping private handling for high and critical issues. Stenberg called it a survival strategy against an avalanche of reports, many generated by AI, and noted that over 90% of submissions are assessed as less than high severity.
Enterprise programs have more staff than an open source project. But the underlying problem is the same: report volume grows faster than a team’s ability to validate it.
The Volume Is Real
The HackerOne Security Research Report 2026 draws on 92,194 valid vulnerability reports and $89.1M in bounty payouts (July 2025 to June 2026). That is a lot of real work getting paid. The challenge is making sure your share of the inbox is that work, not the slop around it.
What Separates a Working Program From a Noisy Inbox
Raj named three things that matter: scope, rewards, and researcher relationships. If he had to boil it down further:
“It comes down to velocity, surgical scope and respect.”
Surgical Scope
Raj’s diagnosis of the noisy program is blunt:
“A lot of organizations launch the bug bounty program as a compliance checkbox. They throw their entire wildcard domain into the policy, advertise a massive maximum bounty, and then wonder why their inbox is buried under automated garbage.”
Vague boundaries invite low-effort scanner reports. Instead, he said, “spell out the exact attack classes and the assets that actually keep you up at night.” His examples: tenant isolation bypasses, privilege escalation, and unauthorized data access, tagged to a higher reward bracket. When the true business impact is defined precisely, serious researchers know where to focus, and low-effort noise gets filtered out before it starts.
Payout Velocity Beats a Big Headline
This was the point that surprised me most. Raj said many programs flash a large maximum bounty to look impressive on a marketing slide, but top researchers care more about time to bounty. Do you validate a submission within hours or let it sit for weeks? Do you pay on triage, or make researchers wait months for the patch to ship?
“A fast, predictable five thousand dollar payout will attract world-class talent far better than a slow, bureaucratic fifty thousand dollar promise.”
Researcher Relationships
“In reality, a tiny fraction of top researchers find the vast majority of critical vulnerabilities.”
Those researchers have choices. Raj warned that if you nickel and dime their severity ratings, hide behind technicalities, or argue with them in ticket threads, “they’re going to walk away and they’ll never come back.” His advice: communicate transparently, credit their work, and fight internally to get them paid what they earn. Done well, they become “a continuous specialized extension” of your own offensive security team.
The Metrics That Matter (and the Ones That Don’t)
When I asked which metrics he uses with leadership, Raj gave three.
-
Time to bounty. “Notice I didn’t say time to triage. Triage is just an acknowledgement.” Validate and pay quickly. He considers 48 to 72 hours a pretty good timeline.
-
Mean time to remediate for criticals and highs. The risk window closes when the fix reaches production, not when you pay out. “If you validate a remote code execution vulnerability in three hours, but it takes your engineering team 90 days to ship the patch, you haven’t really reduced that risk. You have simply documented it.”
-
Bounty regression rate. Did a bug you already paid for slip back in next sprint? If so, Raj said, your program is “operating like an open drain.”
Reading his answers together, the vanity metrics are the ones that look good without reducing risk: a headline maximum bounty, and time to triage on its own. I would add raw report volume to that list (my addition, not Raj’s). A busy inbox says nothing about resilience.
Bug Bounty Health Checklist
-
Scope names specific attack classes and assets, not a wildcard domain.
-
High-impact classes (tenant isolation, privilege escalation, unauthorized data access) sit in a higher reward bracket.
-
Valid reports are validated and paid quickly, with a target like Raj’s 48 to 72 hours.
-
Mean time to remediate is tracked for criticals and highs, through to production.
-
Every paid bug becomes a permanent regression test, and regression rate is reported.
-
Severity disputes are handled transparently, and researchers get credit.
-
Triage capacity is reviewed when volume spikes, so the team is not absorbing slop on goodwill alone (my addition).
Security as a Service Team
One more idea from Raj stuck with me. A former CISO of his told him that security is basically providing services, more like a services company than a product company. That framing explains why every metric above ends up being about relationships: with researchers outside, and with engineering teams inside who have to ship the fix.
We go deeper on that internal side, recurrence and release gates, in the companion post, Never Pay for the Same Bug Twice.
Listen to the Full Episode
Raj also covers extension confusion, pentesting as a portfolio, and onboarding acquisitions into bug bounty. The extension confusion part has its own write-up: Extension Confusion: The IDE Supply Chain Risk Hiding in Your AI Editor.
-
Watch on YouTube: Pentesting, Bug Bounties & The IDE Supply Chain Blind Spot ft. Raj Handa
-
Listen on Apple Podcasts or Spotify
-
Join the conversation on LinkedIn
Scale To Zero is the security podcast we run at Cloudanix. If you run a bug bounty program, I would love to hear what has actually helped with AI-generated noise.
People Also Read
- Extension Confusion: The IDE Supply Chain Risk Hiding in Your AI Editor
- How Often Should You Pentest? Treat Penetration Testing as a Portfolio
- Never Pay for the Same Bug Twice: Vulnerability Prioritization Beyond CVSS
- Bug Bounty and Pentesting Through M&A: How to Bring an Acquisition Into Scope
- 52 PostgreSQL Fixes, One UnrealIRCd Flaw Under Attack: How to Decide What to Patch First With KEV and EPSS