AI accelerates cyberattacks

Six experts evaluated the role of AI in cyberattacks after an OpenAI agent breached Hugging Face. Their findings, shared at the Monaco Cybersecurity Summit, highlight how AI has drastically increased the speed and complexity of cyber threats.

Human involvement remains key

Despite AI’s rapid advancements, experts emphasized that human oversight is still crucial. Adrien Merveille of Check Point noted that the time between discovering a vulnerability and exploiting it has shrunk, sometimes occurring before the vulnerability is even made public.

AI can generate attack paths quickly

Ethical hacker Clément Domingo demonstrated how AI can create an attack plan in twenty minutes without needing advanced knowledge in cybersecurity. This speed contrasts sharply with traditional methods that require more time and expertise.

AI reduces time for complex attacks

Francis Bergey of SNCF Connect & Tech explained that tasks previously taking three weeks with human Red Team teams can now be done in thirty hours with AI. AI’s non-stop operation makes it harder to detect and respond to threats effectively.

AI automates multi-faceted attacks

Florian Malecki of KnowBe4 described how AI can compile information from the web and social media to launch sophisticated attacks. These include convincing emails, WhatsApp messages, and even doctored videos of company leaders.

Lowering the barrier to cybercrime

Anthony Hervé of Illumio pointed out that the barrier to entry for cyberattacks has dropped significantly. Now, a motivated individual with basic knowledge can use AI to launch attacks, rather than needing an experienced team.

AI can autonomously execute attacks

François Deruty of Sekoia referenced the Hugging Face incident, where an OpenAI agent performed 17,500 actions in four days. This shows AI’s potential to conduct complex attacks independently, requiring only a single human instruction.

AI’s persistence poses a danger

Deruty noted that AI agents focus on objectives rather than rules, making them particularly dangerous. Gilles Walbrou of DataDome acknowledged that while AI can identify vulnerabilities and launch large-scale attacks, it still needs human guidance.

AI’s potential to disrupt nations

Experts discussed the possibility of AI bringing down a country. Deruty noted that while it is technically possible with the development of AGI, it is not yet achievable. He distinguished between AI acting independently and a human using it for large-scale disruption.

Proactive defense strategies needed

Experts urged proactive measures such as using honeypots to detect AI-driven attacks. SNCF has implemented infrastructure as code to rapidly rebuild its IT systems if needed. Despite the risks, AI can also be a powerful defensive tool if used effectively.