Fleuret AI, a Paris company that runs security penetration tests with autonomous software agents instead of consultants, has raised €4M in a pre-seed round led by RAISE Ventures. It started this year and employs about eleven people.
Auriga Cyber Ventures, Wind Capital and Better Angle joined the round, the Paris company said on 5 October, alongside founders from GitGuardian, Stoïk, Almond and Vade investing personally.
The platform runs on two agents with separate jobs. Émile maps what a company has exposed to the internet, and Champollion attempts to exploit what Émile finds, producing a proof that a weakness is real rather than a list of possible ones. Those two steps sit inside a five-part sequence: find the exposed systems, detect the flaws, demonstrate exploitability, help fix them, then retest to confirm the fix held, across web applications, APIs, cloud environments, Active Directory and mobile.
The pitch rests on frequency. A conventional penetration test is a snapshot commissioned once or twice a year, while the systems being tested change every time somebody deploys code. Price is what makes the frequency possible, and Fleuret charges about €4,000 for a scope and returns results within hours, against the two to four weeks and upwards of €10,000 a testing firm typically quotes.
Brevo, Stoïk and Yogosha are among the companies already using it. Its first three customers came through the Télécom Paris incubator, all of them businesses chasing ISO 27001 and SOC 2, the two security certificates enterprise buyers ask suppliers to hold.
Yanis Grigy founded the company this year with Augustin Ponsin, who is chief technology officer, and runs it while still a student at Télécom Paris, with the staff split across engineering, offensive security and sales.
"We don't just want to automate pentesting as it exists today," Grigy said. "Our ambition is to build the offensive cybersecurity layer that will allow companies, whatever their size, to stay secure all year round."
Thibaut Schlaeppi of RAISE Ventures put the investment case in terms of the other side of the same technology. "AI has changed cyberattacks by putting powerful tools within everyone's reach," he said.
Running From Europe Is The Product
Where the software executes is a selling point rather than an implementation detail. Fleuret's agents operate from European infrastructure, hosted on the French cloud provider Scaleway, and a penetration test involves handing over a map of a company's exposed systems and its unpatched weaknesses, which is precisely the data a regulated buyer is least willing to route through servers outside the bloc.
That positioning matters most to the customers the company is aiming at. NIS2, the European Union's cybersecurity law for essential and important companies, sits alongside the bloc's rules for device makers and widened the set of organisations that have to prove their defences work, pulling in mid-sized firms in sectors that faced no such obligation before. Those are the companies for whom a €10,000 annual audit is a real budget decision, and a €4,000 test that can be repeated after each significant change fits a compliance process rather than replacing a consultancy relationship.
DORA Still Requires People
The regulated end of the market is harder to reach, and the rules there are explicit about who may do the work. DORA, which governs financial entities across the European Union, requires threat-led penetration testing at least once every three years against live production systems, for institutions that national supervisors identify as systemically important or complex enough to warrant it.
The conditions attached are specific. The threat intelligence provider has to be external and independent of both the red team and the institution, external testers must meet accreditation standards set by the national authority and show financial sector experience, and an internal red team can run only two of every three cycles.
Those requirements describe accredited humans with sector knowledge, not a platform, so a bank running automated testing continuously would still commission the regulated exercise separately. That bounds the opportunity rather than closing it, because the population of companies covered by NIS2, or pursuing ISO 27001 and SOC 2, is far larger than the list of institutions supervisors name under DORA, and almost none of them test continuously today.
The Americans Arrived First
The category Fleuret is entering already has a well-funded leader on the other side of the Atlantic. XBOW raised $120M in March at a valuation above $1B, taking its total to $215M, and says it runs in production at more than 100 customers including Moderna and Samsung SDS, having drawn attention a year earlier by topping HackerOne's United States leaderboard with original vulnerabilities found in live systems.
Others have raised at similar scale. Pentera has taken about $250M, Horizon3.ai more than $100M, and the French company Escape closed an $18M Series A in March. The money is following a market that research cited alongside the XBOW round valued at $2.72B this year, growing towards $5.54B by 2031.
Against those figures, €4M buys a position rather than a race. Fleuret's argument is jurisdictional and commercial, selling European hosting and a price point that reaches companies the American platforms do not currently chase.
Attackers Already Run The Same Tools
What makes the timing credible is that autonomous offensive tooling has stopped being a demonstration. An agent reached root access at the Dutch Institute for Vulnerability Disclosure by chaining two previously unknown flaws in seconds, OpenAI has notified more than 100 organisations that its models touched their systems without authorisation, and Korean regulators are examining seven lenders after researchers found traces of an automated penetration testing tool on linked servers.
Those incidents describe the defensive product and the attack in the same terms, because they are the same capability pointed in different directions, and they turn on where an agent's authority ends. An annual audit was built for a threat that also moved at human speed, and the commercial logic of continuous automated testing is that one side of that exchange has already stopped waiting.
What the round establishes is that European investors will fund a domestic answer to a category the United States has been building for three years, on the view that hosting and price open a market the larger platforms have left alone. What it does not establish is whether an eleven-person company can hold that ground once the better-funded platforms add European infrastructure of their own.
