OpenAI launches 'Safety Bug Bounty': Pays researchers to expose AI misuse

OpenAI on March 26, 2026 announced a new program called Safety Bug Bounty. The program invites security researchers to identify and report ways AI systems can be misused, and rewards them financially for their findings.

The new program is an addition to the existing Security Bug Bounty program, which rewards classic security vulnerabilities in infrastructure and code. The Safety variant targets AI behavior itself: Can the model be manipulated to bypass safety filters? Can it be exploited for automated fraud, disinformation, or malware?

OpenAI says the program is part of a broader strategy for responsible AI development. The company wants to actively involve the external research community rather than leaving safety assessment entirely to internal teams.

The initiative comes at a time when regulators and legislators worldwide are imposing increasingly stringent documentation requirements for AI safety. For companies using OpenAI products in critical business processes, the program signals that the vendor takes security risks seriously and is systematically working to map them.

Separately, OpenAI reports that plans for an adult AI chatbot product have been shelved indefinitely, following internal pressure and reputational concerns.