Beelzebub raises €3M to strengthen enterprise cyber defence with AI

Beelzebub, a cybersecurity startup out of Italy, just closed a €3 million seed round led exclusively by United Ventures. That follows an earlier €300,000 pre-seed backed by strategic investors and advisors, pushing total funding to €3.3 million since founder Mario Candela started the company. Not a huge check by global standards, but enough to signal that someone thinks the AI-versus-AI security fight is worth funding early.

The premise behind Beelzebub is blunt: attackers now use AI to automate the boring parts of hacking. Finding vulnerabilities, writing custom malware, scaling an attack across thousands of targets at once — all of it moves at machine speed now. Meanwhile Europe's regulatory floor keeps rising. The NIS2 Directive and the Cyber Resilience Act are forcing organisations to prove their defenses hold up, not just claim they do.

Instead of betting everything on keeping intruders out, Beelzebub assumes some of them already got in. Its platform leans on three pieces working together. Arcangelo runs simulated attacks continuously to stress-test an organisation's defenses before real attackers do. Beelzebub Managed scatters AI-powered decoy infrastructure through the network to lure and expose anyone poking around where they shouldn't be. And once something malicious turns up, Caronte — described as the platform's AI malware analyst — reverse-engineers the code automatically and writes up incident reports, either in the cloud or entirely on-premises for customers who can't have their data leaving the building.

Candela frames this as a necessary shift rather than a nice-to-have. Human security teams alone, he argues, can't keep pace with attackers who've handed the grunt work to machines. Beelzebub's answer is a platform that keeps adapting itself to new malware types rather than shipping a fixed rulebook. With the fresh capital, the company plans to grow its research team, open commercial offices in Rome and San Francisco before year's end, and push harder into European sales — particularly among organisations that now have to answer to NIS2. It also wants to extend its detection work to a newer target: AI agents themselves.