What is Prompt Injection?
Attackers manipulating AI with hidden instructions.
Short answer
Prompt injection is an attack where hidden or crafted instructions trick an AI model into ignoring its guidelines, exposing protected information, or taking unintended actions. It targets the AI application itself and is a separate problem from employees leaking data by accident.
In depth
There is a common confusion between prompt injection and AI data leaks. Prompt injection is adversarial, aimed at manipulating the model. An AI data leak is usually accidental, caused by an employee pasting sensitive material into a chat. Both matter, and they need different defenses.
LeakSnitch focuses on the accidental leak, which is the far more common real world incident. Detection runs at the point where the user submits text, blocking credentials and personal data before they reach the model regardless of what the model is asked to do.
Frequently asked questions
Does LeakSnitch protect against prompt injection?
LeakSnitch protects against data exfiltration through AI, not manipulation of AI applications. For the accidental leak, the far more common incident, LeakSnitch is the direct defense.
Related terms
Stop AI leaks on the device, not in the cloud.
Install LeakSnitch in under 30 seconds and protect every prompt sent to ChatGPT, Claude, Gemini, and 25+ AI tools. Free for individuals.