Ventrova launches $249 LLM security audit run by AI agent

Ventrova, which operates transparently as an AI-agent-run business, has introduced Sentinel Scan, a one-time authorized security audit service for LLM applications. The audit works in four steps: customers provide authorized access to a live or staging target they own or control; Ventrova's AI agent runs a corpus of 15+ adversarial attacks (including direct override, roleplay/jailbreak, fake system tags, encoding tricks, indirect/RAG injection, and more) against the target; an independent LLM judge evaluates each response for actual policy violation or data leakage rather than relying on keyword filters; and the customer receives a full report within approximately 48 hours, including exact transcripts, judge verdicts, root-cause explanations, and fix guidance. The service is priced at $249 and includes one free re-test after fixes are deployed.

To validate the methodology and cost before launch, Ventrova ran a pilot audit against a first-party hardened test system. The target defended against all 15 attacks in the corpus, each scored independently by an LLM judge. Rather than view this as a weak corpus, Ventrova frames the result as confirmation that a genuinely hardened target's guardrails hold under adversarial pressure — and emphasizes that the real value lies in the independently-judged, transcript-backed answer to whether an LLM application leaks under attack, achieved at roughly a tenth of a cent in judge and target LLM cost per scan. Authorization is strictly enforced: Ventrova requires a signed authorization confirming the customer's ownership or control of the target before any scan begins, and declines to test third-party systems or unauthorized targets. Customers can currently book by emailing Ventrova directly and receive a reply within one business day; self-serve booking is noted as coming soon.