Credentials Detected In AI Usage
Updated: August 11, 2026
Description
Credentials such as API keys, access tokens or webhooks were detected in an employee's AI usage.
Sharing secrets with third-party AI services takes them outside your organisation's control. The credentials may be retained in conversation history, used for model training, or exposed through a breach of the AI provider, allowing attackers to authenticate against your systems and services.
Example Attack
An employee pastes a failing deployment script into an AI chatbot to ask why it isn't working. The script contains a live AWS secret key, which is transmitted to and retained by the AI provider. The key is later exposed and used by an attacker to access the organisation's cloud infrastructure and exfiltrate data.
Remediation
Treat the detected credentials as compromised and rotate them immediately. Review the interactions to determine which services and accounts are affected. Configure guardrails to block secrets in prompts, enforce secret management best practices, and ensure employees understand that credentials must never be shared with AI services.
Security Frameworks
Sensitive fields must be detected and masked, tokenized, or dropped before embedding, and memory and tool writes treated as untrusted input.
Secrets and credentials required at runtime must not be exposed within the model's observable context, including the context window, system prompts, or tool call parameters.
AI interactions, safety filtering and policy decisions, and inference telemetry must be logged to a structured, interoperable schema to support audit and incident response.