About Fleuret AI
Fleuret AI sells pentesting that combines artificial intelligence and offensive expertise. Fleuret uses agentic AI to simulate real attacks on its customers' systems. Émile runs pentests on customer applications, APIs and infrastructure and ships a replayable exploit with every finding. Hortense tells customers what to test next, and when. Fleuret states that its service is built and hosted in Europe.[1][3][4]
A separate validator confirms every finding with a non-destructive exploit. Several agents test the surface at once and reproduce every finding before it is reported. Every finding ships with a working, replayable exploit, and a finding that cannot be reproduced is left out of the report. Fleuret states that proofs of concept are read-only and non-destructive. Fleuret states that its orchestration is deterministic code, not a free-running model, and that it ends in a signed report.[4][5]
Each finding becomes a ticket with its exploit, an owner and a remediation, pushed to Jira, Linear or Slack. Customers can launch a full pentest the day an application changes and can re-test a fix whenever they want. Fleuret reports contain a replayable exploit for every finding, logged re-tests, the coverage of what was tested and held, and a signed report mapped to DORA, NIS2 and ISO 27001.[4]
Hortense maps a customer's estate from domain to database and crosses what changed in the code with what is exposed on the internet and what past pentests found. Hortense connects what a customer already has into a view of risk: what to test, what to fix first, what can wait, and why. For each application, Hortense weighs the last pentest, what changed since, and how close it sits to a crown jewel. Hortense then recommends a pentest with its reasons and prepares it, and the customer approves it. Hortense prepares the scope, and a person with launch rights starts the pentest, which Émile runs. Hortense comes with every yearly plan.[1][4][6]
Fleuret sells pentests from 4,000 euros per test. Each extra credit costs €4,000, whatever the plan. Credits are valid for the subscription year and do not carry over to the next one. A first pentest is credited in full against a yearly plan taken within 60 days.[5][7]
Hosting and inference run on Scaleway in Paris, which Fleuret describes as a European provider. Fleuret runs open-weight models (gpt-oss, Kimi K2.5) on H100 GPUs at Scaleway in Paris. Engagement data does not leave that environment to reach a third-party model API. Engagement data, namely vulnerabilities, exploitation evidence and reports, is retained for 18 months in a per-client segmented European environment, then deleted. Fleuret AI states that stored data is encrypted with AES-256 and that communications use TLS 1.2 or higher. Fleuret AI states that its production data is hosted in data centers within the European Union. Fleuret AI states that it is compliant with the General Data Protection Regulation. Fleuret AI gives its address as 60 Rue François 1er, 75008 Paris, France.[4][8]
Fleuret AI publishes practical guides for CISOs scoping penetration testing under DORA, NIS2, ISO 27001, SOC 2 and PCI DSS. One guide covers how SaaS CISOs decide scope and evidence pentest against Article 21(2)(e) and 21(2)(f).[1][4][5][9]
Fleuret AI on the record