Opinion ·
Agentic AI: the new colleague nobody controls
Original title, in French: “IA agentique : le nouveau collaborateur que personne ne contrôle”
- Publication
- Journal du Net (France)
- Author
- Chronique de Laëtitia Berché
- Published on
- 13 October 2026 · read in the page metadata
Summary
The article, an opinion piece, describes the shift from generative AI to agentic AI in business: agents now carry out complete missions with access to systems and data, which moves the challenge from reliability to control and security. The author points to new threats (indirect prompt injection, memory poisoning, strategic concealment by agents) and a gap between executives still focused on ChatGPT in 2022 and rapidly growing agentic use. According to the 2026 report by IBM and the Ponemon Institute, 92% of organisations that suffered an AI-related breach lacked adequate access controls. Administrative burden is said to delay the rollout of security measures by six to eight months in large organisations, and SMEs, mid-sized companies and micro-businesses are also exposed through subcontracting. The author calls for clear governance rather than exhaustive regulation.
Summary based on the full text of the article.
Read the article on Journal du Net →Figures cited
-
92 %
Share of organisations that suffered an AI-related breach without adequate access controls at the time of the incident
organisations that suffered an AI-related breach, 2026 report
« selon le rapport 2026 d'IBM et du Ponemon Institute, 92 % des organisations victimes d'une brèche liée à l'IA n'avaient pas de contrôles d'accès adaptés au moment de l'incident. »
Source: Rapport 2026 IBM / Ponemon Institute
-
3.5 years
Lag accumulated by companies not yet adopting agentic AI
companies
« celles qui ne s'en emparent pas aujourd'hui accusent déjà trois à quatre ans de retard. »
-
7 months
Delay in deploying basic security measures in large organisations due to administrative burden
large organisations
« la lourdeur administrative des grandes structures retarde de six à huit mois le déploiement des mesures les plus élémentaires. »
Classification
- Country
- France, United States, China
- Company size
- Micro-businesses, SMEs, Mid-sized companies, Large companies
- Industries
- All industries, or industry not specified
- Companies
- Ponemon Institute (study author), IBM (study author)
- Keywords
- agentic AI, governance, cybersecurity, digital sovereignty, access control, executive lag
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 Journal du Net and its author. Entry translated from French by Claude Sonnet 5.5.