EU rules for agentic AI

The AI Act sorts AI by risk

Four tiers: banned, high-risk, transparency duties, and everything else. Most enterprise agents sit in the last two.

• Banned: practices such as social scoring, manipulative techniques and emotion recognition at work. AI Act Art. 5(1)(opens in a new tab)
• High-risk: AI in the areas of Annex III, such as recruitment, managing workers, credit scoring, life and health insurance pricing and critical infrastructure, and AI that is a safety component of products such as machinery. AI Act Art. 6(1)(opens in a new tab)AI Act Art. 6(2)(opens in a new tab)AI Act Annex III(opens in a new tab)
• Transparency: people are told they are dealing with an AI system, unless it is obvious, and generated content is marked. AI Act Art. 50(1)(opens in a new tab)AI Act Art. 50(2)(opens in a new tab)
• Everyone who provides or uses AI supports the AI literacy of their staff. AI Act Art. 4(opens in a new tab)

Is the SRE agent high-risk?

Most likely not, and it is worth knowing exactly why, because a small change of task can change the answer.

• IT operations is not an Annex III area. Critical infrastructure is, but only for safety components, and a system used only for service efficiency, automation or quality control is not a safety component. AI Act Annex III, point 2(opens in a new tab)AI Act Art. 6(1a)(opens in a new tab)
• It becomes high-risk if its failure could endanger health and safety, or if it allocates work to people based on their behaviour or evaluates their performance. AI Act Art. 6(1b)(opens in a new tab)AI Act Annex III, point 4(opens in a new tab)
• If you build it and put it into service under your own name, you are its provider as well as its deployer. AI Act Art. 3(3)(opens in a new tab)AI Act Art. 3(4)(opens in a new tab)

Personal data: the GDPR's questions

Logs, tickets and chats are full of names and e-mail addresses, so the GDPR applies to almost every operations agent.

• What for, and how much? Specified purposes, and no more data than necessary. GDPR Art. 5(1)(b)(opens in a new tab)GDPR Art. 5(1)(c)(opens in a new tab)
• On what basis? Usually legitimate interests, weighed against the rights of the people in the data. GDPR Art. 6(1)(opens in a new tab)
• Is it high-risk? Then a DPIA comes first. GDPR Art. 35(1)(opens in a new tab)
• Who else touches it? A model provider handling data for you is a processor and needs a contract; data leaving the EU needs an adequacy decision or standard contractual clauses. GDPR Art. 28(3)(opens in a new tab)GDPR Art. 45(1)(opens in a new tab)GDPR Art. 46(2)(c)(opens in a new tab)
• Does it decide about people alone? Then those people have rights against solely automated decisions. GDPR Art. 22(opens in a new tab)

The incident clocks

When the agent leaks data or breaks production, several clocks can start at once. Know them before the first incident.

• GDPR: a personal data breach goes to the supervisory authority within 72 hours of becoming aware, where feasible, unless it is unlikely to result in a risk. GDPR Art. 33(1)(opens in a new tab)
• NIS2: early warning within 24 hours, notification within 72 hours, final report within a month. NIS2 Art. 23(4)(opens in a new tab)
• DORA: initial notification within 4 hours of classifying an incident as major and at most 24 hours after becoming aware, intermediate report within 72 hours, final report within a month. DORA Art. 19(4)(opens in a new tab)DORA reporting rules Art. 5(1)(opens in a new tab)
• AI Act: a provider of a high-risk system reports a serious incident within 15 days, or 2 days for a serious disruption of critical infrastructure. AI Act Art. 73(2)(opens in a new tab)AI Act Art. 73(3)(opens in a new tab)

What changed in 2026

The EU simplified parts of the AI Act this year, and has proposed changes to the GDPR that are not law yet.

• The Digital Omnibus on AI, in force since 27 July 2026, moved the high-risk duties to 2 December 2027 (Annex III) and 2 August 2028 (Annex I), and turned AI literacy into a duty to support it rather than guarantee a level. Digital Omnibus on AI(opens in a new tab)AI Act Art. 113(opens in a new tab)AI Act Art. 4(opens in a new tab)
• The Digital Omnibus proposal of 19 November 2025 would change the GDPR, for example to notify only high-risk breaches, within 96 hours. It had not been adopted when this lesson was last reviewed. Digital Omnibus proposal(opens in a new tab)

Classify the agent, map its personal data, name its suppliers, and know your reporting clocks before the first incident, not during it.