Google AI bug bounty program launched, payouts up to PHP 1.75 million

Google has launched a new initiative focused on finding vulnerabilities in its AI systems. This was announced via Google’s Bug Hunters blog with rewards going up to USD 30,000 (~PHP 1.75 million).

The AI bug bounty program is centered around finding security flaws unique to AI technologies, such as ‘prompt injection’. This is an instance where AI could be tricked into performing unauthorized actions.

An example of this is when Google Home unlocks a resident’s front door without notice. It could also be Google Workspace related, like summarizing a user’s emails for the use of a third party.

These incidents are classified by Google as ‘rogue actions’, which the company sees as vulnerabilities in AI harmful to security. Researchers who find security flaws in core Google products can earn from the program.

This includes platforms like Search, Gemini apps, Gmail, Drive, or Workspace. The base reward is USD 20,000, which could go as high as USD 30,000 depending on report quality.

Bugs found in other products like NotebookLM will have lower rewards. In addition, Google said having their AI models ‘hallucinate’ are incorrect outputs and aren’t eligible for rewards.

In addition, participants are encouraged to report issues with harmful or copyrighted AI-generated content through built-in feedback tools.

Here’s a table that outlines the scope of possible issues.