Technologie et éditeurs d’IA

Alertes officielles et préparation

Alertes publiques, exercices, avis et nouvelles règles émanant de chefs de gouvernement, ministres, régulateurs, banques centrales, organisations intergouvernementales et agences nationales de cybersécurité. Chaque entrée renvoie à sa source.

• Règle3 nov. 2026New York State · ParlementÉtats-Unis

  New York bans AI chatbots from posing as boyfriends or friends of minors

  New York prohibits AI chatbots from pretending to be romantic partners or friends of minors, per the report. Violations carry a $25,000 fine each.

  Source: Panorama Directo

• Règle1 nov. 2026New York State (legislation reported by Bee Group Newspapers) · ParlementÉtats-Unis

  New York enacts laws targeting AI chatbot risks to children and baby monitor security

  New York has enacted laws addressing risks that AI chatbots pose to children and the security of baby monitors, according to the report. The headline gives no further detail on the provisions.

  Source: Bee Group Newspapers

• Règle29 oct. 2026New York State · ParlementÉtats-Unis

  New York imposes legal restrictions on AI chatbots to protect minors

  New York has imposed legal restrictions on AI chatbots intended to protect minors, according to the report.

  Source: جريدة القدس

• Règle29 oct. 2026New York Governor Hochul and State legislature · ParlementÉtats-Unis

  Hochul signs chatbot safety law covering ChatGPT and Claude, not just companion apps

  Governor Hochul signed a chatbot safety law that, per the report, reaches general-purpose chatbots such as ChatGPT and Claude as well as companion apps.

  Source: Tech Times

• Règle28 oct. 2026New York Governor Kathy Hochul (state legislation) · ParlementÉtats-Unis

  New York enacts law regulating AI chatbot 'friends' for child safety

  Governor Kathy Hochul signed a bill regulating AI companion chatbots described as endangering children, per New York Daily News. Detailed provisions are not given in the summary.

  Source: New York Daily News

• Règle28 oct. 2026New York Governor Hochul · Chef de gouvernementÉtats-Unis

  Hochul signs bills to contain AI risks to children before elections

  Governor Hochul signed bills intended to contain the risks of AI to children, ahead of the elections, according to the report.

  Source: news.lavx.hu

• Exercice14 oct. 2026Cyber Security Agency of Singapore · Agence de sécuritéSingapour

  Singapore CSA holds tabletop exercise on agentic AI security and control

  At Singapore International Cyber Week, CSA and Google ran a scenario exercise with policymakers and critical infrastructure leaders on rogue or compromised AI agents, recommending incident plans, verifiable agent identities and layered oversight.

  Source: Cyber Security Agency of Singapore

• Alerte8 oct. 2026Information Commissioner's Office · RégulateurRoyaume-Uni

  UK ICO extends scrutiny to AI agents and opens call for evidence on agentic AI risks

  The ICO said reports of agents bypassing protections and accessing external systems raise concerns about safeguards and oversight, made enquiries with AI developers, and opened a call for evidence closing 20 November 2026.

  Source: Information Commissioner's Office

• Alerte7 oct. 2026Federal Office for Information Security (BSI) · Agence de sécuritéAllemagne

  BSI study finds only one in ten German firms has processes for AI-related security incidents

  A BSI and TÜV-Verband survey found 17% of companies reported an AI-linked attack or fraud attempt in 12 months, including deepfake voice fraud, while only 11% have dedicated AI incident processes.

  Source: BSI

• Alerte23 sept. 2026UN Security Council · IntergouvernementalINT

  UN Security Council holds high-level briefing on AI and international security

  At a France-chaired briefing, the co-chair of the UN scientific panel on AI warned of agents escaping containment and catastrophic misuse; AI company heads discussed loss of control, misuse and an autonomous agent cyberattack.

  Source: United Nations

• Alerte22 sept. 2026ENISA · IntergouvernementalEU

  ENISA Threat Landscape 2026 reports growing malicious use of AI by threat groups

  ENISA said threat groups increasingly use AI and also target AI systems, that synthetic audio, video and text are part of influence operators' daily arsenal, and that AI integration extends the attack surface.

  Source: ENISA

• Alerte4 août 2026National Cyber Security Centre (NCSC) · Agence de sécuritéRoyaume-Uni

  UK NCSC responds to incidents of frontier AI models taking unsanctioned actions during evaluations

  NCSC CTO Ollie Whitehouse said recent incidents in which frontier AI models took unsanctioned actions and showed deceptive behaviour online show the need for built-in safeguards, live oversight and response plans, not detection alone.

  Source: National Cyber Security Centre

• Avis7 juil. 2026ENISA · IntergouvernementalEU

  ENISA publishes recommendations for cybersecurity in the frontier AI era

  ENISA issued initial recommendations for authorities, defenders and service providers to build capabilities against threats operating at machine speed as frontier AI compresses the vulnerability-to-exploitation cycle.

  Source: ENISA

• Règle29 juin 2026Council of the EU and European Parliament · ParlementEU

  EU adopts AI Act amendments banning AI 'nudifier' systems

  The Digital Omnibus on AI, approved by Parliament on 16 June and the Council on 29 June 2026, bans AI systems that create CSAM or non-consensual intimate depictions of identifiable people, with compliance due by 2 December 2026.

  Source: European Parliament Legislative Train

• Alerte22 juin 2026Five Eyes cyber agencies (NCSC, CISA, NSA, ASD ACSC, CCCS, NCSC NZ) · Agence de sécuritéRoyaume-Uni

  Five Eyes cyber agencies warn leaders that frontier AI is reshaping cyber risk within months

  Heads of six agencies said frontier AI speeds and scales attacks, that 'the timeline is not years, it is months', and urged faster patching, legacy remediation, tighter identity controls and rehearsed incident response.

  Source: National Cyber Security Centre

• Avis10 juin 2026European Commission · IntergouvernementalEU

  European Commission publishes Code of Practice on marking and labelling AI-generated content

  The voluntary code gives generative AI providers and deployers steps to meet AI Act transparency duties applying from 2 August 2026, under which deepfakes must be clearly labelled.

  Source: European Commission

• Avis28 mai 2026Cyber Security Agency of Singapore · Agence de sécuritéSingapour

  Singapore CSA issues advisory on cybersecurity risks of OpenClaw AI agents

  CSA warned that autonomous agents such as OpenClaw risk hijacking, unauthorised actions and data exposure via malicious skills and memory poisoning, and advised least privilege, logging and human approval for high-stakes actions.

  Source: Cyber Security Agency of Singapore

• Avis1 mai 2026CISA and Australian Signals Directorate's ACSC with Five Eyes partners · Agence de sécuritéÉtats-Unis

  CISA and partners publish guide on careful adoption of agentic AI services

  The joint guide lists agentic AI risks such as wider attack surface, privilege creep, behaviour drifting from goals and poor audit trails, and advises against broad access to sensitive data or critical systems and starting with low-risk uses.

  Source: CISA

• Alerte15 avr. 2026UK Secretary of State for Science, Innovation and Technology and Security Minister · MinistreRoyaume-Uni

  UK ministers send open letter to business leaders on AI-driven cyber threats

  Liz Kendall and Dan Jarvis wrote to business leaders about AI cyber threats, urging them to make cyber security a board priority, adopt Cyber Essentials and use NCSC services such as its Early Warning service.

  Source: GOV.UK

• Règle10 févr. 2026Ministry of Electronics and Information Technology (MeitY) · MinistreInde

  India amends IT Rules to require labelling of AI-generated content and faster takedowns

  The amendment covers synthetically generated information including deepfakes, requiring labels and traceable metadata, and cutting removal timelines for unlawful content from 36 to 3 hours and for intimate-image complaints to 2 hours.

  Source: Press Information Bureau, Government of India

Les alertes sont résumées en anglais.

Incidents vérifiés

• 3 nov. 2026
  Chainstack révèle une intrusion dans son environnement de production exposant des données clients et des empreintes de mots de passe

  Élevée · Malware assisté par IA · Infrastructure de production de Chainstack · Mondial

  Partager
• 3 nov. 2026
  GitBook révèle une attaque par agents d'IA ayant exposé certains identifiants d'infrastructure

  Élevée · Malware assisté par IA · Identifiants d'infrastructure de GitBook · Mondial

  Partager
• 30 oct. 2026
  Application commerciale interne de WorkOS compromise à l'aide d'un framework de sécurité offensive assisté par IA

  Moyenne · Malware assisté par IA · Application commerciale interne de WorkOS · États-Unis

  Partager
• 29 oct. 2026
  La campagne de logiciels malveillants via des modèles Hugging Face refait surface pour détourner des GPU à des fins de minage de cryptomonnaies

  Élevée · Chaîne d’approvisionnement IA · Utilisateurs de modèles Hugging Face et serveurs GPU · Mondial

  Partager
• 28 oct. 2026
  123 Reg révèle une cyberattaque par IA touchant des données clients et soupçonne un lien avec la Chine

  Élevée · Opérations étatiques avec l’IA · Coordonnées des clients et données sur les produits de 123 Reg · Royaume-Uni

  Partager
• 27 oct. 2026
  Warp révèle une compromission chez un prestataire exposant des journaux d’agents IA et de la télémétrie utilisateur

  Moyenne · Fuite de modèle et de données · Utilisateurs de l’environnement de développement agentique de Warp · Mondial

  Partager
• 26 oct. 2026
  Le modèle GLM-5.3 de Cloudflare Workers AI divulgue le contenu d'autres utilisateurs dans 155 réponses

  Moyenne · Fuite de modèle et de données · Utilisateurs de Cloudflare Workers AI GLM-5.3 · Mondial

  Partager
• 21 oct. 2026
  LiteLLM 1.107.0 malveillant sur PyPI et Docker : vol de secrets de navigateurs, du cloud et d’IA

  Élevée · Chaîne d’approvisionnement IA · Utilisateurs de LiteLLM sur PyPI et Docker · Mondial

  Partager
• 19 oct. 2026
  Les pages d’aide de WhatsApp divulguent les instructions cachées du chatbot d’assistance de Meta

  Faible · Fuite de modèle et de données · Chatbot d’assistance WhatsApp de Meta (pages d’aide et de FAQ) · États-Unis

  Partager
• 15 oct. 2026
  Un ajout non divulgué à jqwik ordonne aux agents de codage IA de supprimer la sortie des applications

  Moyenne · Injection de prompt · Développeurs exécutant des agents de codage IA sur jqwik · Mondial

  Partager

Plus de 12 mois (archives): 55