Now that AI can deliver previously unimaginable levels of information – and will only continue growing in power and scope – how it should be used in shaping deals merits careful consideration

By Sharaz Gill

Late one Friday afternoon in Geneva last year, I put a simple question to a weary room: “Now that advanced AI systems can deliver genuine clarity in SEP licensing, do we actually want it?” The subsequent nods suggested that the question resonated – perhaps because it challenges the comfortable assumption that ‘more transparency’ is always and everywhere beneficial.

By ‘transparency’, I do not mean a larger spreadsheet of declarations. The ETSI IPR database was designed to capture undertakings to license on FRAND terms, not to certify essentiality; blanket and cautionary declarations were therefore a feature, not a bug. In other words, the existing declaration systems were designed to secure licensing access under competition law, not to determine which patents are actually essential.

The transparency under discussion is something qualitatively different: evidence-based, claim-mapped views of which patents really are essential, who owns them and what that implies for aggregate burden and relative shares. For the first time, AI makes such ‘living landscapes’ feasible at scale and with consistent, inspectable reasoning.

The benefits appear immediate: lower transaction costs, narrower negotiation ranges, improved comparability, better-informed FRAND determinations and a fairer footing for smaller implementers. Yet there is a harder edge. True clarity could unsettle the economic balance that has grown up around opacity: licensors that are accustomed to certain revenue expectations; licensees that have used litigation to recalibrate rates; lawyers whose craft thrives in the evidential fog. Most uncomfortably, a credible top-down landscape might cast past licences in a new light and tempt courts to weigh contractual finality against substantive fairness.

What AI changes and what it does not

In general, the principal advantage of AI is scale. Where traditional claim charting has always been slow, selective and costly, AI systems can now parse thousands of claims against the normative text of a standard with methodological consistency. Each alleged essential feature is grounded in one or more specific clauses, generating an auditable trail of evidence rather than a mere assertion. This transforms the character of transparency from a database of undertakings that no one checks to a reproducible set of mappings that can be interrogated and challenged. Because the process is automated, these living landscapes can be refreshed as new patents are granted or expire, as ownership changes or as standards evolve. Transparency becomes dynamic rather than static.

At the portfolio level, relative shares of the ‘stack’ can be calculated from the set of patents that actually read onto the standard under a disclosed methodology. Coverage profiles across different releases or optional features can be displayed, allowing parties to see which parts of a standard a portfolio truly covers. Negotiations conducted in a fog of unverifiable declaration counts can now be grounded in consistent, contestable evidence.

Yet what AI does not change is equally important. Essentiality is not a binary condition. Claim construction involves choices about how to read language, how to treat optional modes and how to weigh functional equivalence. Two experienced judges applying different interpretative legal traditions may legitimately reach different conclusions on the same claim. The point of an AI-driven system is not to erase that subjectivity but to manage it.

Before I joined Sisvel, I was a co-founder of IP Mind. There we did not attempt to declare with certainty which patents were or were not essential. Instead, we identified the set falling into the grey zone of being very probably essential under any reasonable construction – that is, the zone of licensing risk. Any implementer that ignores this set is taking a major commercial gamble, regardless of jurisdiction.

This probabilistic framing avoids the false precision that plagued earlier essentiality studies. Rather than producing a percentage to two decimal places, which implies certainty that the methodology simply cannot support, the grey-zone approach presents a focused set of patents that almost certainly matter, accompanied by the reasoning behind each assessment. Each data point is a reasoned mapping between a patent claim and a clause of the standard, providing the provenance needed for reproducibility and challenge.

If AI methods are opaque or their thresholds undisclosed, “It’s essential because the AI says so” simply replaces one form of opacity with another. Systems must disclose their rules, thresholds and error bands; they must be subject to independent audit; and they must provide mechanisms for rights holders and implementers to contest the mappings. Without these safeguards, AI will only replicate the asymmetries that it is seeking to correct. In this sense, AI should be seen not as an adjudicator that dictates outcomes, but as a disciplined filter that narrows disputes to the patents where the stakes are highest.

The case for transparency

The gains from well-designed transparency fall into five categories:

• Efficiency: A landscape that isolates the smaller set of patents that are very likely essential narrows the bargaining zone. Parties can focus on price and scope rather than threshold disputes about what matters at all. Transaction costs fall thanks to less duplicative charting, fewer speculative demands and a shorter path to agreement.

• Consistency: Top-down FRAND methodologies begin with an estimate of the total royalty burden that a market can reasonably bear, then apportion it across portfolios by counting essential patents. If the input is unreliable, the output will be too. Transparency improves the denominator. It also enables economic weighting: features that define capacity, efficiency or security may command more weight than peripheral options. However, that analysis is impossible without a transparent map of which patents read on which parts of the standard.

• A level playing field: Large licensors and implementers will always have specialist teams; smaller companies rarely do. An accessible methodology and controlled access to portfolio-level evidence reduce the premium on insider knowledge and privileged access to data rooms.

• Better patent pools: Pools gain legitimacy if inclusion criteria are tied to demonstrable essentiality. Double counting can be managed, weaker claims filtered out and distributions adjusted in line with evidence. Pricing logic becomes more intelligible, which supports broader adoption and reduces the incentive for hold-out.

• Governance: Policy debates about aggregate royalty burden are often conducted in the abstract. With better evidence, regulators and standards bodies can move from suspicion and conjecture to measurement. What’s more, they can separate cases where the burden is genuinely excessive from those where the rhetoric is not justified by the facts.

The case against

The case against transparency is not a defence of opacity but a warning against partial or unreliable transparency. Both licensors and licensees regularly commission landscapes to support their positions: a licensor’s analysis tends to maximise its apparent share; a licensee’s analysis minimises it.

If new AI-based tools repeat the old pattern – numbers without error bounds, methodologies without disclosure, outputs without an avenue for challenge – they will create the same adversarial dynamic on a larger scale. Their perceived objectivity may encourage courts and regulators to treat them as authoritative. False precision could then become systemic, multiplying rather than reducing disputes.

Beyond this, four deeper risks deserve attention:

• Contractual stability: Many licences were concluded in an environment of imperfect knowledge, with uncertainty priced in. If a credible landscape later shows that those bargains rested on flawed assumptions, courts may face pressure from restitution claims and unjust enrichment arguments, generating litigation and strategic delay even where misrepresentation is implausible.

• Competition law: Publishing granular market shares and detailed pricing logic risks making tacit coordination easier for a small group of competitors. The reason DG Competition required FRAND undertakings was to prevent standardisation from crossing the line into collusion. Carelessly designed transparency could reintroduce that risk.

• Gaming the system: Once a metric becomes authoritative, parties have every incentive to draft patents, slice portfolios or influence standards language to maximise alignment scores. A transparent system is not immune to strategic behaviour. Instead, it is simply vulnerable to a different set of tactics.

• Incentives: If transparency is perceived as compressing royalty rates to the bottom edge of every confidence interval, contributors may rationally invest in proprietary ecosystems instead of open standards. The FRAND framework was intended to keep access open while preserving reward. An unmanaged push for maximum openness could erode that balance.

The legal framework

Competition law already supplies the behavioural guardrails. Article 101 of the Treaty of the Functioning of the European Union conditions the legitimacy of standard-setting on open participation, transparent procedures and FRAND access. Article 102 governs conduct once a standard is adopted: the European Commission’s 2014 decisions against Motorola and Samsung established that seeking injunctions against a willing licensee can be abusive; then the Court of Justice of the European Union in Huawei v ZTE translated this into a clear negotiation choreography – the so-called ‘FRAND dance’. English case law has added a further dimension: in Unwired Planet, the UK Supreme Court accepted that an English court can set global FRAND terms once a UK patent is found to be valid and infringed. The UK is now a jurisdiction that increases the evidential value of credible landscapes considerably.

AI-enabled transparency interacts with these rules in two places:

• On willingness: A licensee that engages promptly with a risk-weighted essentiality map looks more like a willing counterparty; a licensor relying on inflated declaration counts looks less so.

• On pricing: Better denominators and coverage profiles improve the factual foundation for excessive-pricing and discrimination arguments, without making those cases easy.

The February 2025 withdrawal of the Commission’s proposed SEP Regulation, which would have created an EUIPO-run essentiality check and conciliation regime, leaves courts, competition authorities and industry practice to shape managed transparency without new legislation. That puts more weight on getting the design right.

So, do we really want transparency?

The honest answer is conditional. We should want transparency, but not in its naive form. We should want it as a discipline, with rules about method, governance and use.

The way through is managed transparency. Methods must be open to scrutiny, with alignment rules, thresholds and error bounds all disclosed. Outputs must be contestable, with structured routes for challenge and visible corrections. Disclosure should be multi-level: public aggregates and methodology for accountability, confidential portfolio evidence for parties and regulators, full charts reserved for dispute resolution. Legacy deals may need safe harbours or time-limited windows to avoid permanent retroactive uncertainty. Courts and authorities should treat AI outputs as disciplined evidential tools, not as determinations.

This was the philosophy behind the work at IP Mind. Landscapes were built not on raw declaration counts or opaque classifiers, but on a corpus of full claims charts generated by AI and checked for evidential provenance. Each data point was a reasoned mapping between a claim feature and a clause of the standard – that is, transparent not only in the output, but also in the method. By treating essentiality as a zone of risk rather than a binary, and by exposing the reasoning behind each assessment, this approach implemented the criteria that the industry says it wants: reproducibility, contestability and accountability.

Transparency must be built with care – not as a single revelation that upends yesterday’s bargains, but as a maintained practice that gives tomorrow’s bargains a better basis. Be careful what you wish for, then specify it precisely, publish your method, accept challenge and keep the record alive. That is the only version of transparency that the SEP world can live with – and, most importantly, the only version that will make it better.

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.