Data loss prevention and
a prompt injection firewall for GenAI
Where do your code and data go when teams use ChatGPT or Claude? Guardino Sentinel checks each prompt in the browser before it is sent, flags API keys, personal data and injection attempts, and masks what you choose. Detection runs locally; the prompt itself is not sent to Guardino.
> Prompt:
"Debug this prod auth script using AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE..."
The key goes to the AI provider along with the prompt.
> Masked prompt:
"Debug this prod auth script using AWS_ACCESS_KEY_ID=[REDACTED_API_KEY]..."
Flagged before sending and masked in the browser with Ghost Mode. With an account key set, a masked event is logged; the prompt is not.
Mapped to the OWASP Top 10 for LLM Applications (2025)
Rules for instruction overrides, jailbreak role-play, delimiter escapes and indirect injection, with normalisation against hidden Unicode characters and encoded payloads.
Provider fingerprints and entropy checks for API keys, tokens, private keys and connection strings; checksum-validated patterns for card numbers (Luhn), IBAN (mod-97) and Turkish ID numbers (TCKN), plus email and phone.
Flags untrusted input that steers connected tools toward destructive or exfiltrating actions. Enforcing tool permissions stays with the agent host.
Catches requests to reveal the system prompt, and checks model output for canary tokens and verbatim runs of your instructions.
The other categories of the list are on the roadmap. OWASP does not certify or endorse products; this is a coverage map, not a compliance claim.
Status: the browser extension (Chrome, five AI sites) and the LLM gateway are in beta; the REST API and the MCP server run the same detection engines. Pilot deployments are planned with each team.
ChatGPT, Claude, Gemini, DeepSeek, Copilot and OWASP are trademarks of their respective owners, named only to describe compatibility. No affiliation or endorsement is implied.