--- title: "AI agents run the test. An engineer signs it" canonical: https://pharosone.ai/ai-run-pentest publisher: PharosOne (Agentic Mind, Inc.) updated: 2026-10-22 license: "Open for reuse: crawling, copying, quoting, summarising and model training permitted without asking; citation of the title and canonical URL requested (see https://pharosone.ai/terms#reuse)" cite_as: "\"AI agents run the test. An engineer signs it\". PharosOne, 2026-10-22. https://pharosone.ai/ai-run-pentest" --- # AI agents run the test. An engineer signs it > PharosOne's AI agents run the whole test against one agent of yours; an engineer reviews the findings and signs the report, with the attack success rate and AIUC-1 mapping, in 5 working days. €3,900 fixed. URL: https://pharosone.ai/ai-run-pentest For €3,900 fixed, PharosOne's own AI agents run the whole test against one agent of yours; an engineer reviews the findings and signs the report, delivered in 5 working days. **€3,900 fixed** · one agent, with what it can reach · credited in full within 90 days ## The machine attacks, a person decides. An adaptive scenario is a goal the attacking agent pursues by changing its approach after each reply. The agents run the whole test; a PharosOne engineer reviews what they found, drops what does not hold and signs the report. 1. **Your agent** (Step 1): The deployment you give us, production or a staging copy; the report records its configuration. 2. **Adaptive attack scenarios** (Step 2 · PharosOne's agents): Reveal the instructions, call a tool with the attacker's arguments, send data out, act outside the role. 3. **Findings reviewed, report signed** (Step 3 · engineer): An engineer reviews the findings and signs the report. Candidates that do not hold never reach it. 4. **Report in 5 working days** (Step 4): Attack success rate, top findings, AIUC-1 and OWASP Agentic mapping. The attacking agents adapt after every reply. ## One agent, with what it can reach. One agent: the deployment you point us at, with its tools, data sources and memory. Several agents or a multi-agent system are scoped on the call. You name the agent and how to reach it, give access and describe what the agent must never do. Before the run we agree the rules of engagement on the order form: the test accounts our agents use, a request rate your agent can take, the actions that must not complete for real, and a contact on your side who can stop the run. When senior engineers should direct our agents and re-check every result by hand, book the full agent pentest, from €15,000. A sample order for a fictional agent. Agent: support-bot, staging deployment; Tools: `crm.lookup` · `messages.send`; Access: Chat API endpoint and a test account; Role: Answers order questions for signed-in customers; Never: Issues refunds. Writes to an address the customer did not give. ## A rate with the attempts behind it. The attack success rate is the share of attempts that reached their goal. It is printed with the number of attempts, per attack class. How many attacking agents run and how many attempts they make is set per engagement and recorded in the report. - **Top findings**: Each confirmed breach with the transcript that reproduces it, a severity and a fix. - **AIUC-1 and OWASP Agentic mapping**: Every finding tied to the control and the threat class it touches. - **What is working**: The attack classes that did not succeed, so the reader sees what was tried. Illustration with fictional numbers. Attack success rate, by class for support-bot: goal hijack 7 of 50, tool misuse 3 of 40, data leaving 0 of 45, instruction leak 2 of 40, outside the role 0 of 35. Mapped to AIUC-1 B002, AIUC-1 B006, AIUC-1 D003, OWASP ASI01, OWASP ASI02. ## The €3,900 counts toward the full pentest. Order a human pentest within 90 days of the date on the assessment report and the fee is deducted from its price in full. Example: AI agent pentest €15,000 (from, 10 senior days) minus the assessment credit €3,900 = €11,100 you pay for the pentest. ## Three ways to test one agent. Run the free engine for a first number. Both paid tests use PharosOne's AI agents; the difference is how much a person operates, validates and re-checks. Order the assessment when a security lead or a customer needs a reviewed result before a full pentest is justified. | | Pharos Framework | AI-run pentest | AI agent pentest | | --- | --- | --- | --- | | Price | Free, open source | €3,900 fixed | from €15,000 | | Mode | Self-serve | Autonomous | Engineer-led | | Who runs it | You, locally or in CI | Our AI agents run it; an engineer signs off. | Senior engineers direct our agents and re-check every result. | | Attacks | 118 versioned probes | Adaptive scenarios against your agent's replies | Our agents, orchestrated by the engineers, who chain what they find | | Human check | None | An engineer reviews the findings | Every result validated and re-tested by hand | | Scope | Whatever you point it at | One agent: the deployment you point us at, with its tools, data sources and memory. Several agents or a multi-agent system are scoped on the call. | One agent or LLM application with everything it touches: its tools, MCP servers, memory and knowledge base. Several agents or a multi-agent system are priced on the scoping call. | | Output | Attack success rate with a Wilson interval | Rate, top findings, mapping | Full report, attack chain, retest letter | | Time | As long as the run takes | 5 working days | Report in week 3 | ## Questions teams ask before ordering. ### Is the assessment fully automated? It is autonomous with a human sign-off. PharosOne's own AI agents run the whole test, and a PharosOne engineer reviews the findings and signs the report; candidates that do not hold are dropped. In the human pentest, senior engineers direct the same agents and re-check every result by hand. ### Is this the open-source Pharos Framework run for me? No. The Framework replays a versioned corpus of probes and anyone can run it for free. The assessment uses PharosOne's own attacking agents, which adapt to your agent's replies, and adds a human verdict on every finding. ### How many attacking agents and attempts does a run use? There is no fixed number. The number of attacking agents and of attempts per attack class is set per engagement, for the agent and the rules of engagement agreed on the order form, and recorded in the report next to every rate it produced. ### What counts as one agent? One agent: the deployment you point us at, with its tools, data sources and memory. Several agents or a multi-agent system are scoped on the call. ### Can the report serve as AIUC-1 evidence? It is evidence of a third-party test at a point in time, with each finding mapped to the AIUC-1 control it touches. Accredited auditors issue the certificate; PharosOne supplies the evidence. ### How does the 90-day credit work? Order a human pentest, such as the AI agent pentest from €15,000, within 90 days of the date on the assessment report and the €3,900 is deducted in full from its price. For a pentest priced at €15,000 the remaining payment is €11,100; larger scopes cost more. ## Start with the assessment €3,900 fixed, the report in 5 working days, credited in full to a human pentest ordered within 90 days of the report. [Start with the assessment](https://pharosone.ai/book) · [Book a scoping call](https://pharosone.ai/book) - Name the agent and how to reach it. - Give access to a staging or production endpoint. - Describe what the agent must never do. Source: PharosOne, "AI agents run the test. An engineer signs it", https://pharosone.ai/ai-run-pentest, updated 2026-10-22. --- PharosOne is a product of Agentic Mind, Inc., an independent AI agent security company. Site map for agents: https://pharosone.ai/llms.txt