Google Expands Bug Bounty to Address Generative AI Safety

Google expanded its bug bounty program to include generative AI threats

Seeks to incentivize research into AI safety and security

Will categorize and report AI bugs differently

Formed an AI Red Team to find weaknesses in generative AI like ChatGPT

Found LLMs vulnerable to prompt injection and training data extraction

These attacks can generate harmful text or leak sensitive info

Model manipulation and theft also in scope of rewards program

Rewards will vary based on severity, up to $31k for serious flaws