Agentic law
regulationagentic law · law of AI agents
The emerging body of rules and standards on how law reaches an AI agent acting on its own: who answers for what it does, how much authority it can hold, and what evidence it must leave behind.
Agentic law is the still-unfinished conversation about what happens when the party contracting, communicating or deciding is not a person but a software agent. It has three fronts. Accountability: the European framework anchors it in identifiable parties (whoever develops the system and whoever deploys it under their authority), so 'the AI decided' releases nobody. Authority: how far an agent may go alone, with what value and scope limits, and who sets them. Personhood: the open question of whether an agent can hold some legal existence of its own — something being tested in 2026 through proposals for agent-directed corporate entities inside regulatory sandboxes. For a company already running agents, agentic law is not theory: it determines what can be delegated today and what evidence must be kept to prove it.
How it differs from
- Agent governance
- Governance is what the company decides internally. Agentic law is what is imposed from outside.
- AI Act compliance
- The AI Act is one specific regulation in force. Agentic law also covers case law, civil liability and proposals not yet enacted.
Examples
- Deciding who answers when a sales agent promises a discount the company cannot honour.
- Setting the maximum amount an agent can commit without a human signature.
- Keeping the trail of an automated decision so it can be reconstructed for a regulator.
FAQ
- Can an agent sign a contract?
- It can execute the action; the obligation lands on the company that deployed it. So the practical limit is not technical but delegated authority: how much it can commit before a person is required.
- Does this apply if I only run a chatbot?
- Yes, in the cheapest and most immediate part: transparency. An interactive system has to make clear it is AI, and that is already enforceable.