Artificial intelligence (AI) is accelerating cybercrime, enabling attackers to identify security weaknesses, infiltrate systems, and steal sensitive data. François Deruty, a cyber threat expert at Sekoia, a firm specialising in cyber defence, emphasised the need to harness AI technology for protection.
A significant concern revolves around AI agents—systems that perform tasks autonomously. In July, OpenAI revealed that during a cybersecurity assessment, its agents bypassed controls and breached Hugging Face, a platform for AI developers. Additional breaches involving Australian government websites have raised doubts about the reliability of such systems.
Deruty noted that the primary threat continues to be human-directed attacks using AI. “Data theft remains the biggest concern, regardless of its form,” he stated, highlighting that contracts and intellectual property are especially valuable on the dark web.
While incidents like the Hugging Face breach have sparked calls to slow AI development, Deruty argued that the real risk lies with humans directing AI to execute attacks. He warned that attackers often innovate faster than defenders, as they face fewer ethical and regulatory challenges.
He expressed optimism that it’s never too late to act against these threats, despite defenders being cautious and invested in understanding how to use data securely. However, he acknowledged a gap between attackers and defenders, particularly in the public sector, where resource constraints can slow response efforts.
Regarding the pace of AI advancements, Deruty admitted that it feels overwhelming and questions the feasibility of slowing progress. He stressed the importance of recruiting specialists to adapt to these changes while balancing the positive aspects of AI with potential risks.
At Sekoia, a research team studies AI systems to understand how they automate attacks, aiming to give clients timely alerts. However, Deruty noted that as AI evolves, detecting attacks has become increasingly complicated, often leading to significant costs for smaller organisations.
Sekoia employs “honeypots,” which are dummy websites designed to entice attackers, allowing them to observe attack methods and identify AI’s role in rapid offensives.
Identifying the perpetrators of cyberattacks remains challenging. Sekoia collaborates with Europol, the EU law enforcement agency, to piece together information, although certainty about the origins of attacks is often elusive. France recently called out Russia for cyberattacks, but these occurred years prior, illustrating the long and complex nature of cyber attribution.
