Interview ·
"We are not yet seeing fully autonomous attacks emerge": the president of InterCERT France calls for realism in the face of AI labs' marketing
Original title, in French: “« On ne voit pas encore émerger d’attaques complètement autonomes » : le président d’InterCERT France appelle au réalisme face au marketing des laboratoires d’IA”
- Publication
- L'Usine Digitale (France)
- Author
- Alice Vitard
- Published on
- 9 October 2026 · read in the page structured data
Summary
Thibaud Binétruy, president of InterCERT France and head of CSIRT-Suez, plays down the idea of fully autonomous cyberattacks carried out by AI agents, which he says he does not observe in France, while acknowledging AI-"augmented" attacks. He cites the cases of Anthropic (an espionage campaign in which AI reportedly carried out 80 to 90% of tactical operations) and Hugging Face (an intrusion reconstructed as 17,600 actions). He urges organisations to prioritise their cybersecurity investments, roll out multi-factor authentication widely, and apply to the AI agents they deploy the same principles of permission limitation and traceability that ANSSI recommends, rather than focusing on spectacular threat scenarios.
Summary based on the full text of the article.
Read the article on L'Usine Digitale →Figures cited
-
140 teams
Number of member teams of InterCERT France
France
« InterCERT rassemble « 140 équipes » »
Source: InterCERT France
-
1,400 professionals
Number of professionals covered by the InterCERT France network
France
« ce qui correspond à près de « 1400 professionnels » »
Source: InterCERT France
-
85 %
Share of tactical operations carried out by AI in the espionage campaign detected by Anthropic
Campaign detected in mid-September 2025, around 30 organisations targeted
« L’entreprise estime que l’IA a effectué 80 à 90% des opérations tactiques, les humains intervenant à quatre à six points de décision critique. »
Source: Anthropic
-
17,600 actions
Number of actions reconstructed in the intrusion into Hugging Face's systems
Hugging Face incident, agent using OpenAI models, reported on 27 July
« L’entreprise a reconstitué près de 17 600 actions. »
Source: Hugging Face
Classification
- Country
- France, United Kingdom
- Company size
- All sizes
- Industries
- All industries, or industry not specified · Software vendors, IT services companies, telecoms
- Companies
- Agence nationale de la sécurité des systèmes d'information (cited), National Cyber Security Centre (cited), InterCERT France (cited), Hugging Face (cited), Suez (cited), OpenAI (supplier), Anthropic (supplier)
- Keywords
- cybersecurity, ai agents, cyberattack, multi-factor authentication, securing ai agents, ai governance
Entry written by Marsanne Studio from the cited article, with the help of an AI model (claude-sonnet-5). The rights to the article and its illustrations belong to L'Usine Digitale and its author. Entry translated from French by Claude Sonnet 5.5.