A penetration test is a snapshot in time. A team of security consultants spends weeks probing an application, then delivers a report that is often outdated before the ink is dry. Fleuret AI is betting that timeline, and the entire manual workflow, can be collapsed into hours by an AI agent. The Paris-based startup’s platform, named Emile, simulates real attacks against web applications and APIs, aiming to make continuous, audit-ready security validation a repeatable SaaS process [Fleuret AI].
For European companies facing stringent new regulations like the Digital Operational Resilience Act (DORA) and the NIS2 directive, the compliance burden is a powerful forcing function. Fleuret’s wedge is to automate the evidence generation these rules demand. The company announced a €3.5 million financing round in May 2026 to industrialize its approach, comprising €2.8 million in equity led by RAISE Capital with participation from Auriga Cyber Ventures and Wind Capital, plus a €700,000 innovation loan from French public bank Bpifrance [Fleuret AI, May 2026].
The architecture of an autonomous attacker
Fleuret’s technical premise rests on breaking down a traditional pentest into discrete, agent-driven tasks. Instead of a single monolithic scanner, Emile deploys short-lived, specialized AI agents, each targeting a specific attack vector like authentication flaws, injection attacks, or business logic errors [Fleuret AI]. The key differentiator from a vulnerability scanner, the company claims, is that these agents don’t just identify potential issues; they attempt to exploit them in a controlled, production-safe manner to confirm exploitability and generate reproducible proofs of concept [Fleuret AI].
All processing and data storage is hosted on Scaleway’s infrastructure in Paris, a deliberate choice to address data sovereignty concerns for European clients [Fleuret AI]. The platform is built on open-source models, according to a company affiliate, furthering the sovereign cloud narrative [Gabriel MONTEILLARD - Fleuret AI | LinkedIn, 2026].
A compliance workflow, not just a scanner
The product’s commercial appeal is tied directly to regulatory reporting. Fleuret says Emile’s reports map directly to control frameworks like DORA Article 24 and NIS2 Annex I out of the box [Fleuret AI]. The platform includes surfaces for compliance officers, such as direct Jira integration, signed audit PDFs, and board exports, positioning it as a tool for security, technology, and privacy leaders who need to demonstrate continuous validation [Fleuret AI].
Pricing is structured to undercut traditional consultancies. The company cites a cost of €4,000 for a pentest versus €15,000-€30,000 for a firm-led engagement [Fleuret AI]. It also offers wholesale pricing for channel partners, including a 50% discount on its Starter (€5,000) and Growth (€12,500) plans, indicating a planned push through cyber insurers and specialized firms [Fleuret AI].
Building the team behind the agent
Fleuret is an early-stage operation, founded in 2026 and planning to grow from 5 to roughly 12 people by the end of the year [Fleuret AI, May 2026]. The founding team is led by co-CEO Yanis Grigy, who is affiliated with HEC Paris, and co-founder Augustin Ponsin, listed as CTO/COO [VivaTech, June 2026] [Fleuret AI]. A third founder, Macky Dabo, is also listed in association with the company [Shaan Narain - Qonto | LinkedIn, 2026]. The company has already garnered recognition, winning the Jury Prize at VivaTech 2026 [VivaTech 2026 Recap: HEC Paris | HEC Paris, 2026].
Current hiring focuses on core engineering and initial go-to-market roles. The company planned to hire four security or product engineers and one or two sales hires in 2026 [Fleuret AI, May 2026]. An open role for a Full Stack Developer in Paris listed compensation between €45,000 and €75,000 per year [Indeed].
| Role | Name | Title (Source) |
|---|---|---|
| Co-Founder | Yanis Grigy | Co-Founder & CEO [Fleuret AI] |
| Co-Founder | Augustin Ponsin | Co-Founder & CTO/COO [Fleuret AI] |
| Co-Founder | Macky Dabo | Founder [Shaan Narain - Qonto |
The crowded field of automated security
Fleuret enters a competitive landscape populated by established players and well-funded startups. Its identified competitors range from code security specialists like GitGuardian to broader security platforms like Aikido Security and Pentera [Competitors]. The startup’s investor group includes operators from some of these same firms, including Almond, GitGuardian, Stoïk, OVRSEA, and Hornetsecurity, suggesting a strategy to use insider domain expertise [Fleuret AI, May 2026].
The primary risk for Fleuret is proving that its AI agents can match the depth and creativity of human experts across a vast and evolving attack surface. A false sense of security from an incomplete automated test could be more dangerous than no test at all. The company’s answer is its focus on confirmed exploitability and sovereign data handling, betting that compliance-driven buyers will prioritize audit-ready automation over elusive perfection.
Technical breakdown and scale considerations
From an infrastructure perspective, Fleuret’s architecture presents interesting tradeoffs. Using short-lived, specialized agents for each attack vector allows for targeted updates and could reduce the blast radius of a faulty agent. However, orchestrating dozens of these agents to work cohesively across a complex application, without missing chained vulnerabilities that require multiple steps, is a non-trivial systems engineering challenge. The reliance on open-source models hosted on European cloud is a clear market positioning move, but it also means performance and capability are tied to the pace of the open-source ecosystem, not an in-house model team.
The sober assessment for scale is about trust and consistency. Can Emile’s attack success rate hold above a critical threshold as it encounters thousands of unique application architectures? The compliance reporting is a powerful hook, but the platform’s long-term value will be determined by its false-negative rate,the vulnerabilities it misses. At scale, a single missed critical flaw in a high-profile client could undermine the entire automated premise. Fleuret’s next twelve months will be about proving that its agents aren’t just fast, but reliably thorough.
Sources
- [Fleuret AI, May 2026] Fleuret raises €3.5M to industrialize agentic AI pentesting | https://fleuret.ai/news/fleuret-raises-3-5m
- [Fleuret AI] Platform | How Émile Works | Fleuret AI | https://fleuret.ai/platform
- [VivaTech, June 2026] Yanis Grigy - Speakers | https://vivatech.com/speakers/e89295ba-1261-f111-8fcb-6045bd954326
- [Gabriel MONTEILLARD - Fleuret AI | LinkedIn, 2026] Post on LinkedIn | https://fr.linkedin.com/in/pierre-gabriel-berlureau-427320313
- [Shaan Narain - Qonto | LinkedIn, 2026] Post on LinkedIn | https://www.linkedin.com/posts/shaannarain_activity-7090366722926342146
- [VivaTech 2026 Recap: HEC Paris | HEC Paris, 2026] HEC Paris at VivaTech 2026 | https://www.hec.edu/en/news-room/vivatech-2026-recap-hec-paris-heart-europes-most-ambitious-tech-milestone