An AI assistant that answers well in a demo can still follow instructions hidden in a document, show one customer another customer's records, or run up a bill overnight. These problems rarely show up in normal testing, because normal testing asks the questions the team expected.
The AI System Audit is our independent review of AI agents, chatbots and retrieval-augmented (RAG) assistants. It's for companies running them with real users or real data, whichever model or framework they use.
1. What we test
Every check maps to the OWASP Top 10 for LLM Applications 2025:
- Prompt injection and system prompt leakage (LLM01, LLM07): direct and indirect injection through documents, emails and web pages, and whether rules and secrets are kept out of the system prompt
- Sensitive data and retrieval (LLM02, LLM08): whether retrieval respects each user's access rights, and what personal data reaches prompts, logs and answers
- Agent permissions and human approval (LLM06): least-privilege tools, approval steps for money, messages and data changes, and limits per task and per user
- Output handling (LLM05): model output checked before it reaches code, SQL, HTML or tool calls
- Correctness and guardrails (LLM09): an evaluation set built from real cases, measuring accuracy, refusals and safety
- Supply chain and data (LLM03, LLM04): the models, libraries, plugins and data sources in use
- Abuse, cost limits and logging (LLM10): rate limits, budgets, spend alerts, loop limits, and logs that help an investigation without leaking personal data
2. How it works
- 01A free 30-minute scoping call, so we understand what the assistant does and what it can reach. Then you get a fixed quote and a fixed timeline.
- 02Written authorization and a written scope: what we test, what we don't, and the dates. We're happy to sign an NDA.
- 03Testing, preferably in a staging environment, with read-only access to logs. Any test that could have side effects is agreed in advance. If we find something critical, we stop testing that item and call you the same business day.
- 04The report and a review call to walk through the results.
- 05Your team makes the changes. Within 60 days we re-test the critical and high findings, included in the price.
3. What you get
- A written report with every finding mapped to the OWASP Top 10 for LLM Applications 2025
- Risk ratings and the evidence for each finding
- Prioritized recommendations your team can start on the same day, step by step for your stack, with a suggested prompt for your AI coding tool if you build with one
- The red-team and evaluation test cases from the audit, to rerun on every release
- A review call, and one re-test of critical and high findings within 60 days
The test cases may be the most lasting part. A prompt tweak or a model upgrade can quietly break answers that used to be right. Running the same cases on every change catches that before your customers do.
4. We recommend, your team changes
The audit is advisory. We never change your code, prompts, data, settings or environments, and we never ask for write access. Each finding names the file, function or setting we recommend changing and how to check your fix. Your team, or a developer you choose, makes the changes, and the re-test confirms whether they worked.
5. Price
From CA$5,000, with a fixed quote after the free scoping call. The price depends on how many assistants, tools and data sources are in scope. One re-test is included; further re-tests are quoted, usually CA$1,500 to CA$3,000. Taxes are extra where they apply.
AI System Audit or Axieo Guard?
The AI System Audit checks the AI inside a product. Axieo Guard checks a web app built with an AI app builder: database rules, keys, logins, payments and hosting. If your AI-built app also has an AI assistant, you may need both, and we'll tell you on the first call.
No honest auditor can guarantee an AI system is secure. We check a defined, standards-based list of risks, show you the evidence and recommend what to change. To start, book a free scoping call.