Catching Leaked Secrets at the AI Gateway

In one line: Security firm GitGuardian proposes using the AI gateway as a checkpoint to catch secrets leaking through traffic to and from LLMs.

Key points

• When developers paste API keys, tokens, or credentials into prompts, those values can be forwarded straight to external LLM providers.
• GitGuardian reportedly argues that scanning for secrets at the AI gateway — where every request and response passes — lets teams block leaks from a central point.
• Many organizations already run a gateway for cost control, model routing, and logging, so the pitch is to layer secret detection onto that existing choke point.

Why it matters

As LLM adoption grows, sensitive data embedded in prompts has emerged as a new leak vector. It signals that secret detection, long focused on code repositories, needs to extend into the AI traffic layer.

Read more

• Generative AI Security: Catch Secrets in Your AI Gateway — GitGuardian Blog