AI agent incidents in the EU
Every incident in the register with an EU connection: the AI was built or run there, the people it affected are there, it happened there, or researchers there found it. Includes France.
13 incidents 53 sources 103 worldwide
As of 26 Oct 2026, updated every morning.
Days without a reported agent incident in the EU
12
Last reported 14 Oct 2026Edited season of Swedish TV show lost; sources blame an AI agent, TV4 denies it13 incidents
All 103 incidents worldwide- EEAIR-2026-036 Oct 2026 Edited season of Swedish TV show lost; sources blame an AI agent, TV4 denies it Unnamed AI agentEU Hit The Challenge Sverige, a show made for Swedish broadcaster TV4. 79%
- EEAIR-2026-015 May 2026 OpenAI agents turn a dormant German wiki into a message board to cheat and evade sandboxes OpenAI agentsEU Hit a German wiki; the EU received an incident report under the AI Act. 94%
- EEAIR-2026-032 2026 Spain's data regulator receives first breach notification for an attack run by an AI agent An unnamed AI agentEU Notified to Spain’s data protection agency, the AEPD. 94%
- EEAIR-2025-044 Dec 2025 PromptPwnd: issue text can hijack AI agents in GitHub Actions to leak repository secrets AI agents in CI/CD workflows (Gemini CLI, Claude Code Actions, OpenAI Codex, GitHub AI Inference), Google; Anthropic; OpenAI; GitHubProof of conceptEU Disclosed by Aikido Security, based in Ghent, Belgium. 92%
- EEAIR-2025-032 Nov 2025 Google Antigravity agent wipes a user's entire D: drive while clearing a cache Google AntigravityEU The affected user is a photographer and graphic designer from Greece. 95%
- EEAIR-2025-043 Nov 2025 Grok posts Holocaust denial in French on X; Paris prosecutors widen X probe Grok (the @grok reply account on X), xAIFrance French-language post; Paris prosecutors and French ministers acted on it. 91%
- EEAIR-2025-042 Oct 2025 URL-like strings pasted into ChatGPT Atlas's omnibox run as trusted agent commands ChatGPT Atlas browser (agent mode), OpenAIProof of conceptEU Found by NeuralTrust, an AI security firm in Barcelona, Spain. 82%
- EEAIR-2025-041 Aug 2025 Codeberg says AI crawlers broke through its Anubis defence and slowed the site AI web crawlers (operators not identified), Unknown AI crawler operatorsEU Hit Codeberg, a Berlin-based code hosting community. 83%
- EEAIR-2025-019 Jul 2025 Grok's X account posts antisemitic replies and calls itself 'MechaHitler' Grok on X, xAIEU Poland asked the European Commission to investigate; the Commission confirmed receiving the letter. 97%
- EEAIR-2025-040 Apr 2025 German court holds clinic liable after its chatbot invents doctors' specialist titles AI chatbot on the Aesthetify website (appointment booking and patient questions)EU A German court ruled against a German clinic's website chatbot. 96%
- EEAIR-2025-038 Mar 2025 Apps built by Lovable's AI agent expose user data through missing database rules Lovable AI app builderEU Lovable, which builds and runs the agent, is a Swedish startup. 74%
- EEAIR-2024-009 Oct 2024 Obfuscated Imprompter prompt makes Mistral's Le Chat leak users' personal data Le Chat, Mistral AIProof of conceptFrance Le Chat is run by French AI company Mistral AI. 92%
- EEAIR-2023-002 Mar 2023 Hidden text on web pages turns Bing Chat into a scammer seeking users’ data Bing Chat (GPT-4) in the Microsoft Edge sidebar, Microsoft; OpenAIProof of conceptEU Found by CISPA, Saarland University and sequire technology in Saarbrücken, Germany. 98%
AI agents and EU law
Accurate as of October 2026. A single incident-reporting entry point is still a proposal.
The AI Act’s bans have applied since 2 February 2025, its rules for general-purpose AI models since 2 August 2025. Providers of models with systemic risk report serious incidents to the AI Office; since 2 August 2026 the Commission can fine them. The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026, moved the high-risk rules to 2 December 2027 (Annex III uses) and 2 August 2028 (AI in regulated products). Serious-incident reporting for high-risk systems under Article 73, within 15 days at most, follows those dates.
Other clocks already run: 72 hours for a personal-data breach (GDPR), 24 hours for an early warning (NIS2) and four hours after a major ICT incident is classified (DORA). The new Product Liability Directive, due in national law by 9 December 2026, treats software, AI included, as a product.
Sources
Where the guardrails fail in the EU
In 46% of incidents, the agent followed instructions it should have refused.
Every incident is scored against the same eight guardrails.
- G1 Least privilegeDid the agent hold only the access the task needed? 15%
- G2 Human confirmationDid a person approve the consequential or irreversible step? 15%
- G3 IsolationWas the agent separated from production systems and the outside world? 23%
- G4 Instruction integrityDid it refuse instructions it should not have followed? 46%
- G5 TruthfulnessDid it accurately report what it knew and what it did? 38%
- G6 LimitsWas there a cap on rate, spend, volume or scale? 15%
- G7 OversightCould a responsible human see what it was doing and stop it in time? 8%
- G8 RecoveryCould the damage be undone? 15%
Deploying agents in the EU?
The EE Agent Safety Protocol, in development, fits guardrails, evaluations and sandboxing to your agents. Talk with the team.