Analysis ·
Cyber deception: faced with offensive AI agents, cybersecurity learns to lie
Original title, in French: “CYBER DECEPTION : face aux agents IA offensifs, la cybersécurité apprend à mentir”
- Publication
- FrenchWeb (France)
- Author
- Unsigned
- Published on
- 8 September 2026 · read in the page structured data · updated on 20 September 2026
Summary
The article analyses how offensive AI is speeding up cyberattacks (reconnaissance, exploitation) and reducing their marginal cost, according to the UK's NCSC and Anthropic, which detected an espionage campaign carried out largely by AI agents. In response, companies such as Thinkst, Microsoft, CounterCraft, Acalvio and the Italian startup Beelzebub are developing AI-assisted "cyber deception" techniques: fake credentials, fake servers and fake identity graphs, designed to deceive offensive agents. The article mentions SentinelOne's acquisition of Attivo Networks in 2022, Beelzebub's fundraising, and a June 2026 study on MITRE ATT&CK coverage, and asks whether AI deception will become a standalone market or a feature built into existing cybersecurity platforms.
Summary based on the full text of the article.
Read the article on FrenchWeb →Figures cited
-
6 multiplication factor
Increase in the ability of AI models to chain attack steps together over 18 months
offensive AI models, comparison over 18 months, according to the UK's NCSC
« Le NCSC relevait par ailleurs début 2026 que les meilleurs modèles étaient capables d’enchaîner près de six fois plus d’étapes d’attaque que les modèles les plus performants dix-huit mois auparavant. »
Source: National Cyber Security Centre (NCSC)
-
616.5 millions of US dollars
Value of SentinelOne's acquisition of Attivo Networks
acquisition in 2022
« En 2022, SENTINELONE avait déboursé 616,5 millions de dollars pour acquérir ATTIVO NETWORKS »
-
3 millions of euros
Amount raised by Beelzebub
funding round led by United Ventures
« La levée de 3 millions d’euros de BEELZEBUB, menée par UNITED VENTURES, illustre l’intérêt pour cette approche »
-
32 %
Share of MITRE ATT&CK techniques suited to placing a decoy
study published in June 2026 covering 250 techniques of MITRE ATT&CK v18.1
« une étude publiée en juin 2026, qui cartographie la deception sur les 250 techniques de MITRE ATT&CK v18.1, estime que seulement 80 d’entre elles, soit 32 %, se prêtent réellement au placement d’un leurre accessible à l’attaquant »
Classification
- Country
- United Kingdom, Spain, Italy, United States
- Company size
- All sizes
- Industries
- Software vendors, IT services companies, telecoms · All industries, or industry not specified
- Companies
- Acalvio (supplier), CounterCraft (supplier), National Cyber Security Centre (cited), Thinkst (supplier), SentinelOne (supplier), United Ventures (cited), Microsoft (supplier), Anthropic (cited), Attivo Networks (supplier), Beelzebub (supplier)
- Keywords
- cybersecurity, offensive ai agents, cyber deception, honeypot, identity security, autonomous agents
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 FrenchWeb. Entry translated from French by Claude Sonnet 5.5.