Breaking the AI Negotiation Spiral: A Guide for In-house Counsel 

Banner artwork by zb.alexey / Shutterstock.com

We received a demand letter containing dense technical patent allegations. The allegations focused on a product we no longer offer. We classified it as routine and responded quickly. To my surprise, a week later we received completely new allegations; again, long and dense. After some work using AI-aided research, we developed a great response. I expected the other side to back down. Instead, we received adjusted arguments citing different product features. When reviewing the new issues, I realized: this is an AI attack.   

---

Dueling AI is here now 

When I first heard about AI, I pictured the day when parties would negotiate contracts using the technology. I wondered how we’d break the inevitable impasses that arise when two AI agents face each other with extreme standard-playbook provisions. With the patent-related attack, I learned that I’d have to solve the problem now.   

It isn’t only patents or contracts. AI-generated positions can drive any negotiation or dispute. AI buries courts with poorly written briefs, and we’re hearing more about claims from individuals overwhelming government agencies. It is hitting the in-house world, too. I’ve experienced or heard about AI-driven demands in disputes including employees, shareholders, competitors, and third parties across matters like discrimination, fraud, defamation, and contract performance.   

AI-driven negotiation demands can have the appearance of rigor while lacking substance. Opposing sides can develop extensive demand positions quickly, and when you have your own AI to respond with, the process can be much different than unaided negotiations.   

AI-driven negotiation demands can have the appearance of rigor while lacking substance.

As I’ve learned, after you receive an AI-generated demand, quickly firing off your own AI-generated letter isn’t necessarily effective, even with a human-in-the-loop. Responding point-by-point validates the other side’s desire for something to pursue. Giving additional information or explanation provides material for the opposing AI to expand in new directions. A better approach uses techniques like the following to avoid these and other pitfalls: identify the weaknesses that arise from AI overreliance, break the downward spiral that results from unrestrained AI usage, preserve your credibility, and use your own AI effectively.   

Identify the flaws of AI-driven demands 

Review demands for the flaws of AI overreliance. This goes beyond spotting the standardized formatting used by Claude or ChatGPT. You are looking for weaknesses that you can leverage, such as the following patterns.  

Review demands for the flaws of AI overreliance.

Relentless Pursuit of the Demand. Distinct from the allegations and style of the demand itself, a party asserting an AI-driven demand may pursue it relentlessly, irrespective of their demand’s superficiality and errors. Indicators include:   

• Formal written correspondence, where an email or call would be appropriate.  

• Rapid escalation.   

• Pressure to respond quickly to an unnecessarily tight deadline.  

• An extreme claim, often too high.  But can also be excessively low if the party underestimates the situation.   

• Refusal to let anything go, maintaining the claim by shifting through alternative arguments.  

Superficiality. Where a skilled human negotiator prioritizes and focuses on stronger arguments, an AI-driven demand presents any plausible argument it can identify. This results in:   

• Lots of weak theories with no priority given to stronger ones. 

• Unsubstantiated Breadth. Look for unreasonably wide date ranges. 

• Vague causation or injury descriptions. A laundry list of issues presented without any identified causation or injury.   

• Dense arguments that say very little. A complex, formal, and jargon-filled writing style, exhibiting the features common in today’s AI-writing.   

• Template structure vague on details. Recites the claim elements but leaves out substantiation.  

• Unimportant inconsistencies. Raises apparent discrepancies that don’t really matter. This often occurs in response to an explanation or new information.     

Here is an AI-generated example (where no additional substance is provided):  

"Your company’s pattern of incomplete performance, persisting across multiple quarters and customers, was pervasive enough that its effect cannot seriously be disputed. The Company's own shifting account only confirms as much: its March 4 correspondence described 'delivery assurance review' following the Q3 meeting, while its March 18 letter recharacterized the same activity as a 'quality checkpoint.' This evolving narrative, emblematic of a broader institutional pattern of minimization, fundamentally undermines the company’s credibility."   

Errors. AI-driven demands may not be carefully verified and can contain errors. Look for them.  

• Missed Critical Details. AI is limited by access to data. It can miss details simply because it only has access to part of the record. For example, where it relies only on email, it can miss what occurred in person or over the phone, SMS and chat apps.   

• Inconsistent Positions. AI may take inconsistent positions, especially where it develops a response without access to all available correspondence.  

• Hallucinations and misrepresentations. The typical mistakes AI is known to make.  

Break the downward spiral that results from unrestrained AI usage 

Responding to an AI-driven demand requires careful handling to avoid the negotiation spiraling out of control. An effective response manages the level of engagement and information flow and directs the process toward a real conversation.   

Negotiations supported by AI spiral easily. Parties asserting an AI-driven demand arrive with different levels of commitment. Some are not committed and have sent the demand as a fishing expedition — an exploratory letter from a patent assertion entity, for example. Others are upset and fully committed to finding any fault available — a disgruntled business partner or employee. Whatever their initial level of commitment, AI gives them a tool they can use with little effort to craft arguments in their favor. This can drive the relentless pursuit described above: they may read any AI analysis in their favor as validation that they have something to pursue. The situation spirals when the response provides thorough explanations and new information. That becomes fuel for the AI to develop new angles supporting the demand and to build overconfidence. A point-by-point response indicates what has traction and maps the defense. A substantive response to an allegation that was entirely unsupported gives the other side something to work with. Any factual or procedural error in the response suggests a problem and can take center stage. The back-and-forth does not exhaust itself. It continues letter after letter until one side stops feeding it. 

Responding to a relentless AI-driven demand 

When responding to an AI-driven demand, work toward having a genuine discussion about the merits of the dispute. You can do this by pausing to plan your first step carefully, moving the discussion to a genuine conversation, and deciding on a course of action.   

• Pause and strategize. Anything you say may be put under an AI microscope and used against you, so do not attempt to match the formality, volume, and speed of the AI-driven demand. Buy yourself time by acknowledging the demand, setting a later deadline, or taking an extension if needed. Then use the time effectively.  Assess which allegations are credible. Preserve information as appropriate, in anticipation of litigation. Make a plan to achieve closure, which often means asking for specificity, having a call to assess the other side, and then deciding whether to settle or not.   

• Move to a conversation. Guide the negotiations toward a real-time discussion, and resist written exchanges that keep expanding. Seek specificity from the other side first, as discussed below; it sets the table for a conversation on a call. In a real-time scenario, the other side will prepare and naturally prioritize their arguments. It is harder for them to use AI as a crutch. You can ask questions to learn the seriousness of allegations and gauge intent. If they focus on minutiae, fail to prepare, or don’t take the call at all, you learn what you need to know about the quality of their case.  

• Decide on a course. After learning more about the matter, make a decision that gives you finality: settle, mediate, or walk away and prepare to defend. When thinking about mediation, and to some extent litigation, it can be viewed as an extension of the call you had — a live setting that separates the other party from AI dependency. The presence of a neutral decision-maker helps. The benefit of this constrained approach is that it breaks the cycle. It spares you and your team the lost time of chasing a party whose arguments cost nothing to produce and may never be satisfied. Where appropriate, close with a short written record. Explain that you investigated, asked for more information, and discussed it; found that the claim lacks merit; and you will revisit it only if they provide new information. If you weren’t able to investigate due to a lack of specificity, describe what you need.   

Responding to an AI-driven demand’s superficiality and errors  

AI-driven demands can be long, raising a wide range of allegations with varying degrees of strength. Categorize the allegations to stay organized: (a) legitimate with substance; (b) overbroad, unsupported by facts or law, or otherwise weak; and (c) mistaken. Then provide an appropriate response accordingly.  

Engage only legitimate, substantiated allegations. When responding, use your best point and leave the rest aside. Again, any point-by-point response or long explanation that shows your full hand will feed a downward spiral that you want to avoid. If negotiations do continue, you will likely be asked about the same issue from different angles — the same question, asked a different way. The easiest way to handle this is to refer back to your prior answer rather than restate it, since any restatement can be presented as an inconsistency.  

Engage only legitimate, substantiated allegations. When responding, use your best point and leave the rest aside.

You can decline to engage superficial, weak arguments. Other than simply ignoring them, asking for specificity is a good approach. This allows you to frame your position constructively — you require factual details and an actual legal claim to investigate thoroughly.  Only engage when they’ve met a reasonable level of substantiation.   

Handle mistaken allegations and inconsistent positions based on your overall plan for resolving the dispute. Either correct the record early to close the issue, or leave it alone and address it if it comes up later in mediation or litigation.   

Preserve credibility 

Credibility remains critically important when resolving an AI-driven demand. Here are suggestions for winning the credibility battle.  

• The other side’s overreliance on AI. Most courts and mediators naturally find over use of AI as less credible. Let the other side play into this; don't overplay your own hand by explicitly demanding credit for being more careful with AI. 

• Respond with discipline and model reasonableness. Present your position with short, clear, fact-based statements, which can mean avoiding overdependence on AI drafting. Take the claims seriously without feeding the negative negotiation cycle. Avoid name-calling, such as by referring to AI-driven demands as frivolous and baseless.   

• Be careful with process. Missed regulatory deadlines, such as a late response to a privacy request, and other procedural missteps give the other side an easy way to discredit you. 

• Use the record as the case progresses. The early demand correspondence may become evidence in any later mediation or litigation. You can methodically highlight the other side's lack of specificity, its mistakes, and its shifting allegations. Consider the possibility of Rule 11.  

Use your own AI effectively 

The nature of AI-driven demands requires that you use AI as well. Such demands can be too complex and lengthy for human review alone. Here is where AI is a good fit when preparing your response: 

• Issue-spotting analysis. Be sure to provide your AI all applicable data.   

• Litigation-hold analysis.  

• Timeline construction. 

• Research. 

• Verification of your production. AI can check the materials you intend to produce for completeness. This is especially useful where security logs and other complex information is at issue.  

• Don't use AI that subjects you to the issues discussed above.   

• Don’t rely only on AI. Get human help, especially when you are having trouble dealing with complex arguments. Humans can solve problems that AI can’t yet.  

Conclusion 

The patent matter I mentioned was resolved when we stopped feeding new explanations and information, and the parties had a genuine conversation about the merits. While AI changes the volume, complexity, and speed of demands, it ultimately has not changed what resolves them. Parties still have to talk, prioritize, and exercise judgment. This hasn’t changed, and the suggestions in this article are mostly ways to promote that.