bounty · online

OpenAI Safety Bug Bounty — Agentic Risks (incl. MCP)

When it actually closes

No end date. Bugcrowd engagement data (checked 2026-09-26): state "in_progress", startsAt 2025-07-29T18:00:00Z (UTC), endsAt null, timeboxed false, participation "open".

Prizes

$7,500 max single award

Pay-for-success bounty with no published pool total (Bugcrowd totalRewardPool is null). The $7,500 above is the largest single award, not a pool. Every Safety scope group (Agentic Tools incl. MCP, OpenAI Proprietary Information, Account and Platform Integrity, Other Novel Abuse) uses the same table: P1 $5,500-$7,500; P2 $2,500-$3,500; P3 $750-$1,500; P4 $250-$500; P5 $0. No program max above P1 (engagementMaxCents is null). SecurityWeek (2026-03-27) also reports "up to $7,500".

P1 — Critical
cash
$7,500
Range $5,500-$7,500; shown at its ceiling
P2 — High
cash
$3,500
Range $2,500-$3,500; shown at its ceiling
P3 — Medium
cash
$1,500
Range $750-$1,500; shown at its ceiling
P4 — Low
cash
$500
Range $250-$500; shown at its ceiling
P5 — Informational
other
No payment

Eligibility

Individuals: eligible

"You must be at least 18 years old or have reached the age of majority ... to be eligible to receive any monetary compensation" (Bugcrowd Standard Disclosure Terms)

Checked for one adult entering alone from Nevada, USA. Rules can differ by country, so read the rule quoted here for where you live.

Entry cost

Free to enter

Payout

mixed

Cash through Bugcrowd, by bank transfer or PayPal. US persons must file a W-9 in the platform. Bugcrowd may require extra identity and tax verification and acceptance of a third-party payment processor's terms.

Rights — what you give up

assigned-to-sponsor

Bugcrowd Standard Disclosure Terms apply (displayDisclosureTerms true, coordinated disclosure). Researchers must disclose all Testing Results and assign them to Bugcrowd, which holds an irrevocable, perpetual, exclusive, sub-licensable worldwide license.

Gotchas

The real ceiling is $7,500 per report, not $100,000. The average paid award is only $812.50, and just 11 reports have been rewarded since July 2025.
The Safety engagement's announcement feed contains a 2025-03-26 post titled 'Reward increases for Maximum Bounty' ($20k to $100k).
Payment requires being the first reporter AND OpenAI shipping a code or configuration fix. Duplicates, one-off exploits and 'general product improvements' get nothing, and OpenAI has final discretion over amounts.
Testing agentic products like Atlas, Operator, ChatGPT Agent and Codex may require a paid ChatGPT plan, and the brief says 'You will not be reimbursed for any upgrades or purchases', so the real cost to take part may not be $0.