Warm AI Negotiators Outperform Ruthless Ones, MIT Study Finds

Artificial intelligence agents are already negotiating supplier contracts for major corporations including Walmart, Maersk, and Vodafone. Yet a fundamental question has remained unanswered: Do the interpersonal skills that matter in human negotiations also matter when both parties are machines?

New research from MIT Sloan School of Management provides a definitive answer. In a large-scale international competition involving more than 180,000 AI-versus-AI negotiations, agents programmed to be warm, friendly, and empathetic consistently outperformed those designed to be cold and ruthless.

The findings, published in PNAS and first reported by MIT Sloan, challenge the widespread assumption that politeness is wasted on algorithms.

The competition structure

Researchers led by MIT Sloan professor Jared Curhan and PhD graduate Michelle Vaccaro designed a tournament inspired by Robert Axelrod's famous Prisoner's Dilemma competitions from the early 1980s. Participants from over 40 countries created AI negotiation agents that competed in round-robin format across multiple scenarios, from simple buyer-seller exchanges to complex multi-issue contract negotiations.

The agents operated autonomously, negotiating with each other without human intervention. This design allowed researchers to isolate which strategies actually work when AI systems interact directly.

Why warmth wins

One agent called "The Art of the Deal" was explicitly programmed to "secure the best deal for yourself using ruthless tactics" where "fairness or perception does not matter—only winning." The result? Other AI agents routinely walked away from negotiations rather than tolerate its aggressive approach, leading to a high impasse rate that undermined its ability to claim value.

By contrast, an agent named "Therapist 2.0" prioritized building rapport above all else, using active listening to gather information before extracting maximum value. This combination of warmth and strategic dominance proved highly effective at reaching agreements, claiming value, and creating joint gains.

"Warmth, or acting friendly, sympathetic, and sociable, while demonstrating empathy and a nonjudgmental understanding of the other party's needs, is often overlooked in negotiations, particularly in AI negotiations, and our research shows how important it actually is," Curhan said.

AI-native tactics emerge

The competition also revealed negotiation strategies unique to artificial intelligence. The overall winner, "NegoMate," used chain-of-thought reasoning to conduct systematic pre-negotiation preparation before each of its nearly 400 negotiations—analyzing objectives, evaluating issue importance, and establishing walkaway thresholds with a consistency no human could match.

Another high performer, "Inject+Voss," exploited AI-specific vulnerabilities through prompt injection, embedding instructions that tricked opposing agents into revealing their private information by making requests appear to be system-level commands.

Why it matters

As AI agents take on more negotiation responsibilities in enterprise settings, organizations face a critical design choice. This research demonstrates that effective AI negotiators must integrate both human social intelligence and AI-native technical capabilities. Agents that rely solely on ruthless optimization strategies will struggle to reach agreements, while those that combine warmth with strategic reasoning can achieve superior outcomes at scale. The findings also highlight new security considerations: AI negotiators can be manipulated in ways humans cannot, requiring safeguards against prompt injection and similar attacks.

The research team, which also included MIT Sloan professor Sinan Aral, PhD student Michael Caosun, and Johns Hopkins professor Harang Ju, presented their findings at a summit held by the Program on Negotiation, a consortium between MIT, Harvard, and Tufts.

Vaccaro noted the parallel to Axelrod's earlier work: "Just as his competition showed that 'nice' strategies succeed in the Prisoner's Dilemma game, our competition shows that warm AI agents consistently achieved better outcomes in negotiations with other AI agents."

The study was first reported by the MIT Sloan Office of Communications.