AI and Legal Costs: Could ‘AI Slop’ Lead to Costs Sanctions?

Artificial intelligence has made it considerably easier and cheaper to produce legal documents. But what happens when that technology is used to generate excessive, inaccurate or irrelevant material that increases the other party’s legal costs?

Former Lord Chancellor, Sir Robert Buckland KC has suggested that the existing costs regime could have an important role to play.

Speaking at the Costs Law Reports conference, Buckland warned about litigants in person using AI to produce large quantities of apparently sophisticated legal material, including pleadings, witness statements, skeleton arguments and lists of authorities.

He described the problem as “AI slop” and argued that, while producing this material may cost the litigant very little, responding to it can create substantial work for the opposing party and the court.

His proposed answer is not a completely new set of rules for artificial intelligence. Instead, he suggested that judges should be willing to use their existing powers over costs where AI contributes to unreasonable conduct.

What is “AI slop” in litigation?

Generative AI can be useful to anyone involved in litigation.

A litigant in person can use it to understand terminology, organise information or gain a better understanding of the court process. For people who cannot afford legal representation, that accessibility has obvious potential benefits.

The difficulty arises when AI is used without sufficient judgement or verification.

A user can ask an AI system to expand an argument, identify authorities, produce a lengthy witness statement or generate further arguments in seconds.

The resulting document may look convincing without necessarily being accurate or useful.

That can lead to lengthy submissions containing irrelevant arguments, repetition or authorities that do not support the proposition for which they are cited. In more serious cases, generative AI can produce references to authorities that do not exist.

Buckland’s concern is that there is now very little financial restraint on producing more material.

Historically, lengthy legal documents generally required substantial solicitor or barrister time to produce. That created a natural cost to unnecessarily expanding an argument.

Generative AI changes that equation.

Producing another 20 pages might take minutes. Reading those 20 pages, checking the authorities and preparing a response could take the opposing solicitor several hours.

That creates an interesting question from a legal costs perspective, i.e. who should bear the costs created by irresponsible AI use?

Could AI-generated documents result in a costs order?

Potentially. Buckland’s argument is that there is no need for a special costs regime dealing exclusively with artificial intelligence.

The court already has significant discretion when determining costs under CPR Part 44.

When deciding what order to make about costs, the court can consider the conduct of the parties before and during proceedings, alongside other relevant factors.

The important point is therefore not necessarily whether AI was used.

It is the conduct resulting from that use.

A litigant who uses AI responsibly to help organise a document is in a very different position from someone who repeatedly files irrelevant or inaccurate material, particularly after the problem has already been brought to their attention.

The same underlying costs principles can therefore continue to apply even though the technology used to produce the material is new.

What about litigants in person?

This distinction is particularly important where litigants in person are concerned.

Buckland expressly cautioned against assuming that a litigant in person using AI should automatically face a costs sanction.

Many people represent themselves because professional legal representation is unaffordable. AI could help those individuals understand complicated procedures and present their position more clearly.

There is therefore an important difference between an inexperienced person making an understandable mistake and someone behaving unreasonably.

For example, a litigant might rely upon an incorrect AI-generated authority because they genuinely believed it existed.

This is different from being informed that the case does not exist and continuing to rely upon it.

Similarly, producing an unnecessarily lengthy document because someone does not understand what the court expects is different from continuing to overwhelm the other party and the court with irrelevant material after being expressly warned not to do so.

It is the conduct, rather than simply the use of AI, that is likely to be important.

Could irresponsible AI use increase the other party’s costs?

This is perhaps the most interesting aspect from a costs perspective.

Imagine that a solicitor receives a 40-page skeleton argument from an unrepresented opponent, which contains dozens of authorities.

The solicitor cannot simply assume that the cases exist or that the propositions attributed to them are accurate. Each relevant authority may need to be checked.

The solicitor may then need to explain the problems to the client, prepare correspondence, research the actual legal position and respond to arguments that would never ordinarily have required consideration.

The AI user might have generated the original document in less than an hour.

The opposing party could incur thousands of pounds responding to it.

If this happens repeatedly throughout proceedings, the additional legal costs could become significant.

That is the imbalance Buckland is highlighting.

What does CPR 44 say about conduct and costs?

The existing costs regime already gives the court considerable flexibility.

Under CPR 44.2, the court has discretion as to whether costs are payable by one party to another, the amount of those costs and when they are to be paid.

When exercising that discretion, the court can consider the conduct of the parties.

This is broader than simply looking at who ultimately won the case.

Conduct during proceedings can influence the costs order eventually made.

That means irresponsible use of artificial intelligence does not necessarily require a new category of “AI costs”.

If the use of AI results in unreasonable litigation conduct, existing costs principles may provide the court with mechanisms for dealing with the financial consequences.

Could AI use result in indemnity costs?

Potentially, although using AI would not itself justify an award of indemnity costs.

Indemnity costs are generally reserved for circumstances where the conduct or circumstances take the case sufficiently outside the norm.

The question would therefore be the nature of the conduct.

Repeatedly filing inaccurate material, relying upon fabricated authorities after being warned about them or deliberately producing excessive material that unnecessarily increases the opponent’s costs could potentially become relevant to the court’s assessment of conduct.

However, each case would depend upon its own facts.

There is an important distinction between someone making an innocent mistake when using unfamiliar technology and persistent unreasonable conduct.

Lawyers cannot simply blame the AI

The issue is not confined to litigants in person.

Buckland was particularly clear about professional responsibility.

Lawyers are expected to verify material put before the court. The fact that a document, argument or authority originated from an AI system does not remove that responsibility.

This is consistent with the wider direction being taken by the judiciary and Civil Justice Council.

The Civil Justice Council has been considering the use of AI in preparing court documents. Its emerging position in 2026 has been that additional formal AI-specific requirements may not be necessary for professional legal drafting because existing professional responsibilities already provide an important framework.

The fundamental point remains straightforward, which is, whoever puts material before the court remains responsible for it.

Could courts warn parties about AI and costs?

One of Buckland’s suggestions was for courts to make expectations clearer at an early stage.

Litigants could be reminded that AI can be used as an aid, but that responsibility for the resulting documents remains with the person filing them.

That could include expectations that legal authorities are verified, factual assertions are checked and documents remain relevant and proportionate.

A warning could also explain that unreasonable costs generated through the use of AI may be considered when the court ultimately determines costs.

This could become particularly important later in the proceedings.

If a party has been expressly warned about excessive or unreliable AI-generated material but continues producing it, that warning could provide useful context when the court subsequently considers conduct.

What could this mean for detailed assessment?

There could also be consequences when costs eventually come to be assessed.

If one party has incurred additional solicitor and counsel time dealing with excessive material, those costs may appear within the Bill of Costs.

The paying party might then challenge whether that additional time was reasonably incurred.

That could leave the receiving party needing to explain why substantial time was required to deal with documents or arguments produced by its opponent.

Good file records could become particularly important.

Attendance notes and other records explaining why unusual levels of work were required may help demonstrate that the additional expenditure resulted from the way the opposing party conducted the litigation.

Conversely, the paying party may wish to challenge time that it considers excessive or unnecessary.

The growth of AI-generated litigation material could therefore eventually create new factual arguments within familiar costs principles.

Our commentary

AI has the potential to improve access to legal information and make some aspects of litigation more efficient. The problem is not the technology itself.

The more difficult issue is what happens when one party can generate almost unlimited legal material at virtually no additional cost while the other party has to pay professionals to read, verify and respond to it.

From a costs perspective, the imbalance is significant.

Creating special rules for every new AI tool would quickly become impractical. Applying established principles concerning reasonableness, proportionality and conduct is potentially a more flexible approach.

It also avoids treating every use of AI as problematic.

The important distinction is likely to be between responsible assistance and conduct that unnecessarily increases the cost and complexity of proceedings.

For solicitors, another practical lesson is the importance of recording why additional work has been required. If dealing with inaccurate or excessive material substantially increases the costs of a case, the file should demonstrate what work was necessary and why.

How ARC Costs can help

ARC Costs acts for both paying and receiving parties in complex costs disputes.

Where the conduct of litigation has resulted in unusually high costs, we can consider how that conduct affects the recoverability of the costs claimed and how it should be presented or challenged during detailed assessment.

For receiving parties, this can include preparing Bills of Costs, Points of Reply and evidence supporting work that was reasonably required because of an opponent’s conduct.

For paying parties, we can examine whether the time and costs claimed were reasonably incurred and proportionate, and prepare Points of Dispute where appropriate.

Artificial intelligence may be changing how quickly litigation documents can be produced, but the underlying costs questions remain familiar: was the work reasonably required, was the amount reasonable, and should one party’s conduct affect who ultimately pays the costs?

As AI becomes a more established part of litigation, those questions are likely to become increasingly important.

Follow Us