In cases where there are application security vulnerabilities, threat actors can exploit them to inject harmful code into an AI agent’s responses, causing the agent to provide skewed or manipulated information.

This manipulation could lead to deceptive outcomes, posing a potential risk to users who rely on the AI agent’s outputs.

With many public AI tools available across the internet, we should be vigilant and where possible only use AI tools from trusted organizations with strong security processes in place.