AI Governance: Managing Autonomous Contracting Risks

We now live in a brave new world where AI agents can negotiate, finalize and execute a contract without any human intervention.  That may be great from a convenience perspective, but the potential downsides are enough to make you lose a lot of sleep. This Debevoise blog addresses the risks associated with AI agents’ ability to enter into binding contracts, including the uncertain availability of traditional contract defenses, and offers guidance on how to mitigate them.

In order to manage these risks, the blog identifies the key components of an AI governance framework for autonomous contracting. This excerpt addresses the need to make the limits of the agent’s authority clear to contractual counterparties:

Make key limits visible to counterparties. Because limits that a counterparty never sees may not protect the company, companies should consider building key limits into their contractual relationships. Options include addressing agent use and authority in master and/or form agreements (UETA generally allows parties to vary its rules by agreement, and the UNCITRAL Model Law defers to attribution procedures the parties agree on), disclosing when an agent is negotiating and what it cannot agree to, and providing that an agent’s communications on higher-risk matters are not binding until confirmed by a designated human. Disclosure may also strengthen a mistake defense by giving the counterparty reason to know when an agent has exceeded its limits. Companies should also review their agreements with AI vendors, whose liability caps and disclaimers may leave the company bearing the full cost of an agent’s error, and consider whether their insurance would cover those losses.

Other elements of a governance framework identified by the blog include assigning clear accountability, defining and technically enforcing the AI agent’s contracting authority, ensuring appropriate governance practices for the underlying knowledge (such as contract templates, precedent libraries, and negotiation playbooks), establishing independent monitoring and auditability, and preserving meaningful human expertise and intervention.