--- title: "Penetration testing for companies that run AI agents" canonical: https://pharosone.ai/ 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: "\"Penetration testing for companies that run AI agents\". PharosOne, 2026-10-22. https://pharosone.ai/" --- # Penetration testing for companies that run AI agents > Senior-led penetration testing of your infrastructure and the AI agents on it, from €9,500 with the report in week 3 and a 90-day retest included. Pharos Radar reads your agents' sessions in production and stops an attack before it lands. URL: https://pharosone.ai/ Senior engineers and our own AI agents test your infrastructure and the agents on it, at a fixed price. Pharos Radar watches your agents at runtime and stops an attack before it lands. Pentest from €9,500 · AI-run pentest €3,900 · report in week 3 · retest included. - Published CVEs: 22 by our team - Speaking: Black Hat speaker · DEF CON 28 IoT CTF winner - Experience: 15+ yrs in cyber security - Report: Week 3 with a live debrief for your team - Retest: 90 days included, with a closure letter ## Three products for the same agent Both pentests use our own AI agents. In the AI-run pentest they run the whole test and an engineer signs off; in the human pentest senior engineers direct them and re-check every result. Radar watches the agent at runtime, and all three land in one cabinet. ### Human pentest · from €9,500 Senior engineers test the stack and the agents. **Engineer-led.** Senior engineers direct our agents and re-check every result. Fixed price, report in week 3, retest within 90 days. [Scopes and prices](https://pharosone.ai/pentest). ### AI-run pentest · €3,900 AI agents run the test. An engineer signs it. **Autonomous.** Our AI agents run it; an engineer signs off. Adaptive attack scenarios against your agent, report in five working days, credited in full to a human pentest ordered within 90 days of the report. [Start with the assessment](https://pharosone.ai/ai-run-pentest). ### Runtime guard · Pharos Radar · from €1,500 / mo Radar stops the attack at runtime. **Per-turn integration stops a flagged call in the same call**; the one-line wrapper acts from the next turn. Radar is not a proxy between your application and the model provider. [How the guard works](https://pharosone.ai/runtime-guard). ## Every finding lands in one evidence set Human pentest findings, AI-run findings and runtime flags land in the same cabinet and map to the same controls, so the evidence set keeps growing with every test and every session. - Human pentest: Senior engineers, report in week 3. - AI-run pentest: Our agents, report in five days. - Runtime guard: Radar, in production. - One cabinet: findings and flags, mapped to the same controls. - One evidence set: SOC 2 · DORA · ISO 27001 · AIUC-1 · ISO 42001. Of the 100 AI security vendors we reviewed in October 2026, two sell automated testing, a human red team, a pentest and compliance evidence together; PharosOne adds the runtime guard to that set. ## Radar monitors every session and detects the attack Radar reads the conversations and tool calls your integration sends, and detects manipulation, data leaving and actions outside the agent's role. With per-turn integration the verdict returns in the same call, so a flagged reply is held and a flagged tool call does not run. With the one-line wrapper the verdict arrives as the call returns and applies from the next turn. Every detection lands in the cabinet with its AIUC-1 control. Inputs reach the agent; each tool call your integration sends passes through Radar. Trusted calls run. Flagged calls are held and never execute. - Turns and tool calls you send - Per-turn: held in the same call - Findings mapped to AIUC-1 - SaaS or on-prem licence Details: [Runtime guard: Pharos Radar](https://pharosone.ai/runtime-guard). Research preview: A probe on the agent's hidden states. Numbers and pre-registrations on [research](https://pharosone.ai/research). ## Fixed prices, published, with the days behind them Start small or start with the full test. Every human pentest includes scoping, hands-on testing by senior engineers, the report, a live debrief and one retest within 90 days with a closure letter. Prices are net of VAT; the exact number is fixed in the SOW after the scoping call. ### Where to start - Step 1 · free: [Pharos Framework](https://pharosone.ai/knowledge/pharos-framework). The open-source engine, 118 probes, runs locally or in CI. - Step 2 · AI-run pentest: [€3,900](https://pharosone.ai/ai-run-pentest). AI agent security assessment. One agent, with what it can reach. Our AI agents run it; an engineer signs off. A report with the attack success rate and AIUC-1 mapping in 5 working days. Credited in full to a human pentest ordered within 90 days of the report. - Step 3 · human pentest of the agent: [from €15,000](https://pharosone.ai/pentest). Senior engineers direct our agents and re-check every result. Report in week 3, retest included. - Step 4 · runtime guard: [from €1,500/mo](https://pharosone.ai/runtime-guard). Pharos Radar. Every production session read, flagged calls held; findings with their AIUC-1 control. Billed yearly. - Step 5 · continuous testing: [from €3,000/mo](https://pharosone.ai/knowledge/continuous-testing). Agent changes tested before release, at the volume fixed in the SOW; quarterly retests as AIUC-1 evidence. Billed yearly. ### Pentest scopes **Web apps · APIs · cloud · source code**: from €9,500, usually €12,000–18,000 for an app with several roles and its API. - 7+ senior days; the engineer who scoped it tests it - One app and its API, up to 3 roles, about 75 endpoints; more in the SOW - Report in week 3, live debrief, one retest within 90 days - Mapped to SOC 2 CC4.1 · PCI DSS 11.4 · ISO 27001 A.8 · DORA Art. 24–25 **AI agents and LLM applications**: from €15,000, one agent or LLM app with everything it touches: tools, MCP servers, memory, knowledge base. - 10 senior days; senior engineers direct our AI agents, validate every result and chain what the agents find - Every bypass reproduced and confirmed by a person - Report in week 3, one retest within 90 days; quarterly retests on the platform as an option - Mapped to AIUC-1 · ISO 42001 · EU AI Act Art. 9 and 15 · OWASP Agentic Top 10 **Internal network**: from €12,000, usually €12,000–20,000. - 8–9 senior days from an employee seat - Up to about 200 hosts and one AD domain - Report in week 3, one retest within 90 days - Written for PCI DSS 11.4 and DORA reviewers [All scopes and prices: WAF and EDR bypass from €9,500, red team from €38,000](https://pharosone.ai/pricing). ## Questions buyers ask before the call ### Will my auditor accept the report? It follows PTES and OWASP WSTG and maps to SOC 2 CC4.1, PCI DSS 11.4, DORA Art. 24–25 and ISO 27001 A.8. The SOW names the framework you face. ### Does Radar stop an attack or only report it? Both. With per-turn integration the verdict returns in the same call, so a flagged reply is held and a flagged tool call does not run. With the one-line wrapper the verdict arrives as the call returns and applies from the next turn. Either way the session lands in the cabinet as a finding with its AIUC-1 control. Radar is not a proxy between your application and the model provider; with per-turn integration your application waits for the verdict before it releases a reply or runs a tool, so it must handle a timeout. ### Can we try anything before a call? Yes. Run the free Pharos Framework on your agent with no API key and no network: npx skills add pharosone/pharosone. For Radar, wrap one agent at app.pharosone.ai and the first sessions appear in the cabinet. ### Where does our data go? The engine runs inside your perimeter, and your keys and transcripts stay there. Radar runs as SaaS at app.pharosone.ai, with the judge model on PharosOne’s own servers and no outside inference provider, or under an on-prem licence on your servers, where transcripts never leave your perimeter. For certification evidence we upload signed statistics only. More: [FAQ](https://pharosone.ai/faq). ## Next step Book a 30-minute scoping call with the engineer who runs the test: [book a call](https://pharosone.ai/book), or write to audit@pharosone.ai. The SOW with a fixed scope and price follows within three days. - One field to start: work email. The calendar asks the rest. - Not ready? Ask for the redacted sample report. - Already running agents? Wrap one with Radar and see the first sessions come in. - Building agents? Run the open-source engine. Source: PharosOne, "Penetration testing for companies that run AI agents", https://pharosone.ai/, 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