AI agent liability follows delegated authority and human authorization

As AI agents increasingly execute financial trades and manage funds without constant human intervention, legal experts and technical frameworks are addressing the complexities of liability and authorization.

Brickken CEO Edwin Mata argues that because AI is not recognized as a legal person under current law, liability must follow the authority granted to the software rather than the software itself. He suggests that responsibility should be determined by investigating who authorized the agent and what powers were delegated. This perspective is supported by the fact that AI acts as a technical system on behalf of a natural or legal person.

In the European Union, technical guidelines suggest that while AI agents can draft contracts and initiate signing workflows, they should not be treated as qualified electronic signatories under eIDAS regulations. Instead, a two-phase process is recommended where an agent proposes an artifact, but an authorized natural person must review and approve the payload to maintain a valid legal signature.

Reports from Sandmark indicate that existing laws do not provide a single answer for losses caused by autonomous agents, meaning courts may apply contract law, negligence, or product liability depending on whether the loss resulted from authorized strategies or faulty design and weak safeguards.