Negotiating with algorithms
Contract automation is shifting from support tool to negotiating force, raising a central governance question: Who is accountable when AI shapes the terms? Kellie Blyth of Addleshaw Goddard shares her insights.
Artificial intelligence is rapidly moving from the back office to the negotiating table.
For many businesses, the first wave of generative AI in legal and commercial functions was relatively contained. Tools helped draft clauses, summarise documents, compare contracts, or accelerate research – while the human professional remained visibly in control.
That boundary is becoming less clear.
AI systems can now analyse contractual terms, identify risks, recommend negotiating positions, suggest amendments, and interact with counterparties via email, chat, and contracting platforms. As capabilities become more sophisticated, AI agents are likely to play an increasingly active role in shaping the terms on which businesses transact.
The attraction is obvious. Legal teams are under pressure to manage growing workloads while controlling costs; and technology can perform in seconds tasks that once consumed hours. But the greater the role AI plays in influencing a commercial outcome, the more important another question becomes: If an AI-generated decision results in an unfavourable term, financial loss or contractual dispute – who is responsible?
For the moment, the answer is relatively straightforward.
AI MAY NEGOTIATE, BUT IT CANNOT TAKE RESPONSIBILITY
Under current legal frameworks, AI is generally treated as a tool rather than a legal person with independent rights and obligations. The fact that increasingly sophisticated technology can appear to reason, recommend, or even negotiate does not transfer responsibility from the organisation deploying it.
That distinction matters.
A company cannot simply point to an AI system and argue that “the technology made the decision”. Businesses are expected to evaluate the tools they deploy, understand the risks and establish appropriate policies, procedures and employee training.
There may also be contractual routes through which responsibility can be allocated to technology vendors. But the AI itself does not become accountable simply because it played a material role in reaching the outcome. Accountability remains with people and organisations.
The analogy with other technologies is useful. A car may allow somebody to travel from A to B far more quickly, but the driver remains responsible for using it appropriately. AI presents a much more sophisticated version of the same underlying principle.
The complication is that businesses may increasingly give these systems greater autonomy.
If an authorised system communicates a position, accepts a term or participates in a process that concludes a contract, the organisation may find itself bound by the outcome. Questions around authority, contract formation and agency therefore need to be considered before autonomous capability is deployed, not after something goes wrong.
EFFICIENCY CREATES A NEW CATEGORY OF RISK
An AI system might drop an important instruction, focus disproportionately on one legal issue, or fail to understand the broader commercial context of a negotiation. An apparently minor drafting error in a template could then be reproduced across hundreds or thousands of agreements.
This is one of the paradoxes of AI: The same scale that creates extraordinary efficiency can amplify mistakes.
The legal profession has already seen warnings of what can happen when credible AI outputs are accepted without adequate scrutiny. Judicial guidance in England and Wales has highlighted the risks posed by AI “hallucinations” that generate incorrect or misleading information and reinforced the continuing personal responsibility of judicial office holders for material produced in their name.
Contract negotiation presents a different context, but the lesson is transferable. AI can generate something that looks polished, logical, and authoritative while still being wrong.
A negotiating position can also be legally sound in isolation while being commercially nonsensical. An experienced lawyer may know that winning a particular drafting point is not worth jeopardising a strategic relationship, delaying a transaction or conceding something more valuable elsewhere. AI does not necessarily see that bigger picture.
That is why human judgement remains so important.
THE CONTRACT BEHIND THE AI MATTERS TOO
Businesses also need to understand the commercial framework governing the AI they use.
An enterprise AI solution may sit atop of a stack of different technologies, and rely on cloud services or products operated by third parties. This creates a chain of contracts, with the terms available to the ultimate customer in part dictated or restricted by the terms applied by the service providers much further down the chain.
The terms governing most off-the-shelf generative AI tools make clear that outputs may be inaccurate and place responsibility on users to assess whether those outputs are appropriate for their intended purpose.
For low-risk administrative tasks, that allocation of responsibility may be commercially manageable. The tolerable level of risk is likely to be different when AI is being asked to analyse a high-value transaction, influence material contractual negotiations, or perform another business-critical function.
Larger and more sophisticated customers are therefore increasingly likely to push for service-specific commitments such as warranties linked to performance, clearer obligations around outcomes, and more nuanced allocations of liability.
We have seen a similar evolution before with cloud contracting. Terms that initially appeared non-negotiable changed as major customers demanded minimum reasonable levels of protection and those expectations gradually filtered through the market.
AI contracting is likely to evolve in much the same way.
REGULATION IS MOVING, BUT NOT AT THE SAME SPEED EVERYWHERE
Globally, a recognisable set of principles has emerged around responsible AI, including fairness, accountability, transparency and explainability, safety and governance, and mechanisms through which outcomes can be challenged. The Organisation for Economic Co-operation and Development (OECD) AI Principles, first adopted in 2019 and updated in 2024, have become an important international reference point for trustworthy AI. They emphasise the key role of human oversight, transparency, safety, accountability, and traceability amongst other principles.
The European Union has gone furthest in creating a comprehensive legislative framework. The EU AI Act follows a risk-based model, with high-risk models having been prohibited in 2025, transparency obligations for certain AI systems applicable from 2 August 2026, and further requirements to be applied in due course under the Act’s phased implementation timetable.
The UAE, meanwhile, has pursued an innovation-friendly approach while placing responsible deployment at the heart of its AI ambitions. Its international AI policy is built around six principles – advancement, collaboration, community, ethics, sustainability and safety – and supports alignment with international standards and cooperation on AI governance.
However, when it comes specifically to AI participating in commercial contracting, there is currently limited UAE judicial authority illustrating how these issues will be treated in practice. That makes the contractual framework particularly important. Questions around AI performance and outcomes are likely to turn principally on the contractual allocation of obligations, limitations of liability, and the available remedies when things go wrong.
In the majority of cases, established legal principles will continue to apply: contract law, data protection, consumer protection, financial regulation, employment obligations, directors’ duties, and broader corporate governance, depending on the context.
The novelty of the technology alone neither removes those considerations nor makes them any less relevant.
BUSINESSES NEED TO DECIDE HOW FAR THEY ARE PREPARED TO LET AI GO
The practical question for boards and general counsel is therefore not simply whether their organisation should use AI. In many businesses, that debate has already been overtaken by reality.
The more important questions are: Where is AI being used? What is it authorised to do? And where does human supervision and responsibility for its outputs begin and end?
Organisations should map AI use across the business and distinguish between fundamentally different activities. Asking a system to summarise a contract is not the same as allowing it to recommend a negotiating position. Allowing it to recommend a position is not the same as authorising it to communicate that position to a counterparty. And that is different again from permitting a system to independently accept contractual terms.
Those boundaries need to be deliberate.
Businesses should establish clear escalation points and determine which decisions always require human sign-off. For material contracts and negotiations, meaningful human oversight should remain central rather than becoming a rubber-stamping exercise after the AI has effectively made the decision.
Employees also need to understand how those boundaries impact their role in practice. Policies sitting unread on an intranet will provide little protection if employees do not know what tools they may use, what information they may input, what outputs require verification and when they must escalate.
As always, good governance must be fully operationalised, not just theoretical.
AUDITABILITY WILL BECOME INCREASINGLY IMPORTANT
Businesses should also consider what happens when a decision is challenged months or even years later.
If AI influenced the negotiating position, can the organisation establish from its records what instructions were given to the system, what it produced, what changes a human made, and who ultimately approved the decision?
Logs and audit trails of prompts, outputs and human decisions may become increasingly important in disputes, investigations and internal reviews. Specifically, these records serve as an important tool for managing downstream legal risk and auditing compliance with internal risk policies.
This approach is also consistent with the OECD’s accountability principle, which emphasises traceability in relation to datasets, processes and decisions across the AI decision-making lifecycle.
This is also where governance and commercial efficiency begin to converge.
As AI pricing models evolve, businesses may become more disciplined about determining where the technology genuinely creates value. Not every task requires the use of an AI engine, and greater visibility over usage and cost, may encourage organisations to deploy it more sparingly in future, only where the return justifies both the expense.
ACCOUNTABILITY IS A FEATURE, NOT A BARRIER
None of this should be interpreted as an argument against AI-enabled negotiation.
Used well, AI has the potential to transform contracting. Accelerating reviews, identifying inconsistencies, improving access to preliminary information and freeing lawyers and commercial teams to concentrate on strategic concerns and higher-value judgement calls.
Well-governed organisations may gain a significant competitive advantage from harnessing those capabilities.
But speed should be tempered by appropriate restrictions on authority, and apparent sophistication should not be mistaken for genuine ability.
For boards and general counsel, the priority should therefore be to understand where AI has been adopted, establish clear governance and human oversight, scrutinise vendor contracts, maintain appropriate audit trails and continue monitoring regulatory and market developments. Those are not obstacles to innovation; they are what makes sustainable innovation possible.
AI may increasingly draft the clause, recommend the compromise and perhaps even communicate with the counterparty. What it cannot currently do is carry the legal and commercial responsibility for the result.
Text by:
Kellie Blyth, partner, Commercial (Technology and Data), Addleshaw Goddard
