AI and the future of (F)RAND negotiations

For the first time, both licensors and licensees can realistically examine all the patents in a portfolio rather than just a selected sample. This represents a fundamental change in the information dynamics of SEP licensing

By Sharaz Gill

For decades, SEP licensing negotiations were shaped by a simple practical constraint: comprehensive essentiality analysis was expensive. While licensors could justify the cost of claim charting because the resulting work product was reusable across multiple licensing campaigns, implementers often had to deal with numerous licensors across a wide variety of standards and could seldom justify conducting exhaustive reviews of every asserted portfolio.

That model may now be changing. There is emerging evidence that implementers are starting to respond to licensing overtures with AI-generated analyses of large numbers of patents. Rather than responding solely to a licensor’s selected examples (so-called ‘proud lists’), they are deploying AI systems to generate competing claim charts of their own, portfolio landscapes and alternative views of essentiality.

AI has not eliminated disagreement over essentiality. Questions about claim construction, technical interpretation and legal judgement remain; and reasonable experts can still reach different conclusions. What AI has changed are the economics. Large-scale SEP evaluation that once required teams of engineers and patent attorneys can now be performed far more quickly and at a fraction of the cost, transforming the availability of evidence in both licensing and litigation.

Today, for the first time ever, both sides in a negotiation can potentially look at an entire portfolio, as opposed to just a selected sample. Consequently, we are likely to see a fundamental shift in the information dynamics of SEP licensing.

In a recent Sisvel Insights article, I argued that AI-generated patent landscapes should be judged not only by their conclusions, but also by the transparency, reproducibility and contestability of the methodologies that produced them. This article explores the next stage of that evolution. What happens when both sides of a (F)RAND negotiation gain access to those same capabilities and begin generating competing analyses at scale?

The traditional (F)RAND negotiation model

SEP licensing has always been conducted under conditions of uncertainty. Parties know that some patents are likely to be essential, but rarely know precisely how many, how important they are or how a court might ultimately assess them.

For implementers, comprehensive essentiality analysis was often prohibitively expensive. Patent owners, by contrast, could justify substantial investment in portfolio-wide review because the resulting work product could be reused across multiple licensing negotiations. This economic asymmetry made direct evidence scarce, particularly on the implementer side.

As a result, negotiations evolved around indirect proxies. Patent owners relied on metrics such as declaration counts, pool evaluations, litigation outcomes and their own portfolio reviews. Implementers typically relied on their own selective technical reviews, validity challenges, public litigation records and whatever market information was available to them. Both sides understood the limitations of these indicators but accepted that they provided a workable basis for reaching agreement.

This shaped the structure of (F)RAND negotiations. Discussions frequently centred on patent proud lists rather than a comprehensive portfolio analysis. Existing licences carried significant weight because they were among the few observable market signals. Litigation often functioned as a mechanism for obtaining information that could not be accessed elsewhere, including access to those licences.

The system was far from perfect, but it functioned surprisingly well. Parties negotiated within broad ranges of uncertainty and accepted that complete precision was neither achievable nor necessary.

The industrialisation of analysis

The most important contribution of AI is not that it performs essentiality analysis better than human experts. It usually does not. Its significance rather lies in scale.

For decades, the limiting factor in portfolio evaluation was the sheer volume of work required to compare thousands of patent claims against thousands of pages of technical specifications. However, modern AI systems can screen portfolios, identify potentially relevant sections of standards and generate first-pass mappings with an accuracy and at a speed that would have been inconceivable only a few years ago.

This capability is not confined to licensors. Implementers, litigation funders, patent pools, courts and independent evaluators have increasing access to similar tools. Human expertise remains essential, but experts can now focus their attention on the patents that genuinely merit detailed review.

The impact is analogous to electronic discovery in litigation. Computers did not replace lawyers; they changed the economics of identifying the documents worthy of legal attention. AI is beginning to perform a similar function for SEP analysis.

For the first time, comprehensive portfolio analysis is becoming economically feasible for both sides of a negotiation. The consequences of that shift are just beginning to emerge.

When both sides have AI: mutually assured confusion

One of the earliest consequences of AI is the emergence of what might be called ‘counter-charting’.

Traditionally, claim charts flowed in one direction. Patent owners presented charts as evidence of portfolio strength and implementers challenged them. AI is altering this dynamic. Implementers appear to be increasingly conducting portfolio-wide reviews of their own and generating competing analyses at a scale that would previously have been impractical.

The implementer’s position is therefore shifting from, “Show us that you have a significant number of SEPs,” to “We have analysed your portfolio ourselves and we think that you are overstating your portfolio strength.”

In principle, this should improve the quality of discussions. Competing analyses expose assumptions, reveal methodological differences and encourage closer examination of disputed patents. Yet this development also introduces a new challenge.

In practice, licensors are unlikely to be able to rely solely on manual review when faced with large volumes of AI-generated counter-positions. Attorney and engineer time is a precious resource in the licensing world and deploying it to assess large volumes of AI-generated claim charts is simply impractical. It is likely, therefore, that licensors will increasingly utilise AI systems themselves to assess, rebut and prioritise implementor challenges. The result is a self-reinforcing cycle in which each side uses AI to respond to AI-generated analysis from the other.

This seems akin to an analytical arms race. As the cost of generating claim charts and counter-charts falls, both sides become able to produce technical evidence at an unprecedented scale. The result may not be greater certainty, but rather a growing volume of competing analyses that neither side can evaluate quickly or economically.

This is where the information overload begins. The technology reduces the cost of producing technical arguments for both parties simultaneously. Claim charts generate counter-charts. Counter-charts generate rebuttals. Rebuttals generate further responses. The volume of analysis can expand far more rapidly than the capacity of human experts to evaluate it.

The future (F)RAND negotiation is therefore likely to involve several competing essentiality narratives, each supported by an extensive body of apparently reasoned technical analysis. In a world where both sides can generate technical analysis at an industrial scale, the risk is not a shortage of evidence, but an excess of it.

Evidence: can there be too much of a good thing?

At first sight, AI-driven transparency appears to be entirely beneficial. Intuitively, more information should produce better decisions. Yet history suggests that technological advances often create new constraints as well as eliminating old ones.

AI may present an analogous challenge for SEP licensing. AI can generate useful evidence, but it also generates vastly more evidence. The costs of producing a plausible claim chart are falling dramatically, while the costs of evaluating that chart manually remain largely unchanged.

One can already observe the emergence of what might be termed ‘information flooding’. Hundreds of challenges can be generated at minimal cost. Some may be credible. Many may not. Yet the recipient must still review them.

This does not imply bad faith. Many implementers will use AI precisely as intended: to better understand the portfolio that they are being asked to license. The practical effect, however, is that negotiations increasingly become debates about analyses of patents rather than patents themselves.

AI, (F)RAND conduct and willingness

The implications of AI extend beyond portfolio analysis. They may also influence how courts assess the conduct of parties engaged in (F)RAND negotiations. Following Huawei v ZTE, courts have increasingly focused on willingness and good-faith engagement as indicators of (F)RAND-compliant conduct.

AI complicates this picture. An implementer entering into negotiations with a portfolio-wide analysis and a detailed set of technical objections may argue, with some force, that it has engaged seriously with the asserted rights. At the same time, though, volume alone must not be allowed to become a substitute for genuine engagement. One of the emerging challenges for courts may be distinguishing between legitimate technical disagreement and the strategic deployment of AI to increase the cost and complexity of negotiations.

A key question therefore arises: if an implementer presents hundreds of AI-generated objections, is a licensor obliged to respond to each one individually? Requiring licensors to rebut every AI-generated challenge could create powerful incentives for procedural delay. Ignoring such analyses altogether may be equally problematic.

Courts faced with this question are likely to look beyond the quantity of material exchanged and examine whether the positions taken reflect genuine technical reasoning or whether they represent the kind of bulk production that the information-flooding dynamic makes possible at negligible cost. Courts may therefore decide to focus on methodology rather than outputs alone. If a party wishes a court or counterparty to rely upon AI-generated analyses, it could reasonably be expected to explain how those analyses were produced, what assumptions were applied and how the results were validated.

In summary, in an AI-enabled environment, willingness may depend not only upon participation but also upon whether the evidence presented is sufficiently transparent and reproducible to permit meaningful scrutiny.

AI-mediated evidence review

If AI can generate competing claim charts, can it also be used to compare them?

One possible response to evidential overload is the development of AI-mediated review systems that can identify points of disagreement and highlight the issues that genuinely require expert judgement.

Such systems would not determine legal essentiality or (F)RAND terms. Rather, they would perform a form of evidential triage, reducing thousands of pages of competing analyses to a manageable set of genuinely disputed issues.

Again, any such system would require transparent methodologies, reproducible outputs and independent oversight. Nevertheless, if AI is transforming the production of evidence, the SEP ecosystem may eventually require equally sophisticated tools for evaluating it.

What happens to patent pools?

Given my current role, it would be remiss of me not to consider the effects of this phenomenon on patent pools. Patent pools occupy an unusual position within the SEP ecosystem. Their credibility rests on the fact that the patents they administer have been independently evaluated by human technical experts prior to inclusion.

AI places that foundation under challenge. As noted above, as implementers gain access to portfolio-scale analytical tools, they can increasingly generate competing essentiality assessments at relatively low cost. Pools may therefore find themselves defending established evaluation records against large volumes of AI-generated counter-analysis.

The appropriate response is likely to be a greater emphasis on methodology and transparency. Independent human evaluation, conducted against defined criteria by qualified experts, remains a form of high-quality provenance that AI-generated analysis cannot yet equal. At the same time, pools may increasingly use AI to enhance and scale their own review processes, occupying an important middle ground between bilateral negotiations and fully independent institutional review.

The next competitive advantage

Much of the current discussion focuses on claim chart generation. Yet once a capability becomes widely available, it rarely remains a durable source of competitive advantage. Claim charting itself is likely to become increasingly commoditised. Future differentiation may lie in methodology, provenance, reproducibility, confidence assessment and dispute management.

The organisations that succeed will not necessarily be those that produce the largest volumes of claim charts; they will be those whose analyses can withstand scrutiny from courts, counterparties and independent reviewers, and indeed extensive AI reviews. This may create opportunities for independent evaluators whose role is not to generate analysis, but to assess competing analyses.

The changing thicket challenge

For decades, SEP licensing operated in a world where comprehensive analysis was scarce and information asymmetries were inevitable. AI is changing that. Portfolio-scale analysis is becoming available to both sides of a negotiation and evidence that was once difficult to obtain can now be generated at unprecedented scale.

The history of technology offers a useful lesson. Progress rarely removes constraints altogether. More often, it relocates them. Therefore, we should not be surprised if AI reduces the patent thicket only to create a new one: a claim chart thicket, in which competing analyses proliferate faster than they can be meaningfully evaluated. Whether the institutions, practices and negotiation models of the SEP ecosystem are ready for that transition remains an open question.

The challenge will no longer lie in obtaining evidence; it will lie in evaluating competing bodies of evidence, understanding the assumptions on which they rest and determining which analyses are sufficiently reliable to inform licensing, litigation and valuation decisions.

The next chapter of SEP licensing is unlikely to be defined by who can generate the most claim charts. It will be defined by who can establish the most credible methods for assessing them.

Yet every new source of friction creates an opportunity for innovation. If AI succeeds in creating a claim chart thicket, perhaps one of the entrepreneurs reading this article will find a way through it.

Sharaz Gill is Head of Portfolio Management at Sisvel

The opinions expressed within this article are the author’s and do not necessarily reflect the views of Sisvel. The content is for informational purposes and should not be taken as legal advice.