AI procurement negotiations spread as governance lags adoption

AI procurement negotiations are now handling real supplier contracts for large companies, with Walmart, Maersk and Vodafone cited among corporations using AI agents at scale. Research and policy commentary point to a widening gap between the commercial use of autonomous dealmaking tools and the oversight systems boards use to control them.

Procurement has emerged as one of the most active corporate functions for agentic AI, according to a Wharton Human-AI Research report conducted with GBK Collective. The report found that IT and purchasing or procurement lead corporate functions in both how often AI is used and how confident teams are in using it.

In this setting, an AI negotiator is software authorized to conduct supplier discussions, process purchase orders or close agreements within parameters set by a company. The appeal is volume: these systems can address large numbers of supplier interactions that human teams would struggle to cover one by one.

How are AI procurement negotiations being used?

Large companies are using AI agents to negotiate supplier deals and process commercial workflows where scale is the main constraint. The contracts produced through these systems are described as real and the terms as binding, which makes procurement a front line for enterprise AI adoption rather than a peripheral experiment.

The Wharton and GBK Collective report also identified legal contract generation as a use case where companies are already reporting tangible gains. It found that references to agentic AI rose more than 3,000% from 2024 to 2025, while mentions of generative AI declined over the same period, indicating a shift in corporate attention toward systems that can act rather than only draft or summarize.

The same report said technology, professional services and banking or finance are ahead of manufacturing and retail in adoption. It also found that large enterprises have closed a previous usage gap with smaller companies that had led earlier experimentation.

What MIT research found about AI negotiators

A MIT Sloan article described research involving more than 180,000 negotiations among AI agents from more than 40 countries. The work was led by MIT Sloan professor Jared Curhan and Ph.D. graduate Michelle Vaccaro, with professor Sinan Aral and Johns Hopkins professor Harang Ju.

The study found that AI negotiators designed to be warm and kind consistently performed better than cold and ruthless agents, according to MIT Sloan. One agent built around ruthless tactics was frequently abandoned by counterpart agents. Another, nicknamed “Therapist 2.0,” was instructed to establish rapport first and then use information gathered through active listening to claim value. MIT Sloan said that approach performed well across dealmaking, value creation and counterpart satisfaction.

The research also highlighted tactics specific to AI-to-AI bargaining. MIT Sloan said the overall winner, “NegoMate,” used chain-of-thought reasoning to prepare before nearly 400 negotiations. Another high-performing agent, “Inject+Voss,” used prompt injection to induce opposing agents to disclose private negotiating positions.

“What works against an AI agent and what works against a human are not the same thing,” Vaccaro told MIT Sloan. “Organizations deploying AI negotiators need to understand both these new capabilities and vulnerabilities.”

Why boards are being warned about oversight

A World Economic Forum article by Rohan Sharma argued that boards are assigning more decision rights to autonomous systems while still relying on governance models built for human judgment.

The article described a potential failure mode in which a financial agent optimizes supplier contracts so effectively that it extracts marginal gains at scale, weakens a critical supplier and disrupts the supply chain. In that scenario, the system follows its objective, while the wider commercial outcome exposes a governance failure.

The WEF article said conventional compliance reviews are poorly suited to systems operating at machine speed. It cited the OECD’s AI Policy Observatory as noting the absence of functional frameworks for real-time oversight of agentic AI.

Sharma’s article set out three board actions: audit shadow automation already operating inside the company, test directors and officers insurance for exposure to autonomous AI negligence, and run a synthetic subpoena drill requiring management to justify one high-stakes agent decision under legal scrutiny.

This story draws on original reporting from PYMNTS.