"FRAND FIRST" Meets AI: What Munich's New Guidelines Mean for SEP Implementers

Munich's 7th Civil Chamber wants parties to address the commercial merits of an SEP dispute earlier and more directly. Its new FRAND Guidelines offer an early "FRAND FIRST" hearing, with briefs limited to 25 pages plus exhibits. Comparable licences come first. Top-down analysis remains an important cross-check and may take a larger role where suitable comparables are missing. [1] [2]

For implementers, this is both a risk and an opportunity. A low counteroffer is not enough. Its assumptions must be explained and supported by evidence at an early stage. Otherwise, the implementer may still face an injunction.

AI can make that level of preparation possible across an entire portfolio. Recent judicial developments in the UK and Australia already point in this direction. AI can organise, compare and synthesise large amounts of evidence, while the underlying sources remain traceable and lawyers and experts remain responsible for the conclusions. [7]–[11] SEP portfolios are an obvious field for this approach.

Put together, these developments point to a new model for FRAND disputes. AI performs the broad, source-linked analysis. Lawyers and experts test the difficult points and shape the legal and commercial position. The court receives a short and focused submission backed by a much deeper evidence base.

The central proposition is simple: the 25-page brief should be the interface, not the whole analysis.

Key points at a glance

Issue	Practical implication
Status	Current position of Munich's 7th Civil Chamber only. Not a judgment or binding rule; it does not bind other chambers, appeal courts or the UPC.
FRAND FIRST	Either party may request an early FRAND hearing. Briefs may run to 25 pages plus exhibits.
Willingness	If liability to pay is accepted, the Chamber expects a non-refundable payment on account based on the implementer's offer. Additional security may be required.
Valuation	Comparable licences come first. Top-down is generally a cross-check, but may become primary if suitable comparables are unavailable.
Main risk	Even payment and security may not suffice if a FRAND offer remains rejected and the lower counter-position is unpersuasive.
AI opportunity	First narrow the portfolio by product relevance. Then use AI-assisted claim-level analysis on the remaining patents and defend the findings that matter.

1. How the 7th Civil Chamber proposes to approach FRAND

The Guidelines are a detailed position paper dated 13 August 2026 from the 7th Civil Chamber of the Munich I Regional Court, chaired by Presiding Judge Dr Oliver Schön. [1] [2] They are important in practice, but they are not legislation, a judgment or binding guidance for other courts.

Their most visible proposal is "FRAND FIRST". Either party may ask for an early hearing devoted to FRAND, supported by a brief of no more than 25 pages plus exhibits. The direction is clear: identify the decisive commercial issues early and present them systematically. [1] [2]

The substantive hierarchy is also clear. The Chamber starts with sufficiently comparable licences from the relevant licensor. A top-down calculation generally tests the result. It can take a larger role where no suitable comparable or established licensing practice exists. [1] [2]

For implementers, valuation is only part of the task. Where the obligation to pay is accepted and only the amount is disputed, the Chamber expects a non-refundable payment on account. Its amount is normally based on the implementer's own offer. Depending on the absolute and percentage gap between the offers, additional security may be required. [2] [3]

Without adequate payment or security, the Chamber says it will generally not need to examine the SEP holder's offer. In case 7 O 5007/25, it nevertheless examined the offers in the alternative. [3] Formal compliance is therefore important, but not necessarily decisive on its own.

This is why FRAND FIRST can front-load injunction risk. It does not merely invite an early academic discussion about rates. The implementer may need to show, at the same early stage, that its counteroffer is serious, economically coherent and supported by the available evidence.

2. Why a lower counteroffer needs more than a lower number

Munich's recent case law gives the warning in concrete terms.

In case 7 O 7655/25 (Avago vs. Renault), the implementer had paid the amount it considered undisputed and had provided security. The Chamber considered this sufficient to show outwardly that no obvious holdout was taking place. It nevertheless found that the implementer lacked the necessary internal willingness because it continued to insist on a substantially lower rate based on an expert report that the Chamber found unconvincing. The court granted injunctive relief. [4]

The lesson is not that an implementer must move towards the SEP holder's demand merely to appear willing. It is that a substantially lower counteroffer should be capable of explanation. What product base was used? Which standards and releases matter? How was the licensor's share calculated? Were applications and expired patents treated consistently? What supports an adjustment for portfolio quality or product relevance?

The importance of party evidence is visible in the opposite direction. In ZTE v Samsung, case 7 O 64/25, the Chamber used Samsung's party-expert estimate that ZTE held 5.7% of the overall mobile-communications standard, including applications, as one input in its top-down analysis. [5] The decision does not approve AI or essentiality filtering, and it does not validate the report as a whole. But it does show that party analysis can affect the inputs used by the court. Both this judgment and case 7 O 7655/25 are first-instance decisions under appeal.

The practical conclusion is clear: a lower counteroffer needs an evidentiary foundation. That evidence must be available early, address the relevant commercial and technical assumptions, and withstand challenge. It must also be combined with the payment, security and negotiating conduct required under the Chamber's approach.

3. Samsung v ZTE identifies the scale problem

The 2026 English FRAND judgment in Samsung v ZTE explains why this is difficult. The licensing experts agreed that claim charts and technical discussions are the most reliable way to evaluate portfolio quality. The court also found the essentiality studies discussed in that case prone to inconsistent results and cherry-picking. Yet it recognised the practical limit: "This cannot be done in the context of a FRAND trial such as the present for practical reasons." [6]

That leaves a familiar gap. Courts and negotiators value claim-level evidence, but traditional review is difficult to scale across hundreds or thousands of patents. Samples can be attacked as unrepresentative. Generic counts are scalable, but may say little about actual essentiality, validity or product relevance — the gap between declared and genuinely essential patents is exactly where the argument sits. A product-relevance filter can reduce the universe first, but it does not decide essentiality or validity. The patents that remain still require the much heavier claim-level work.

Samsung v ZTE therefore identifies a missing capability, not a permanent limit. Courts value claim-level analysis, but a conventional manual review cannot cover an entire portfolio within a FRAND trial. AI can move much of that work to an earlier stage. It can test a far broader part of the portfolio before the parties select the issues and patents that genuinely matter for negotiation or litigation.

The purpose is not to let a model determine the FRAND rate. It is to give the commercial and legal decision-makers a broader and more consistent evidence base from which to develop and defend their positions.

4. Courts are beginning to design the framework for AI-assisted litigation

AI is already moving into court proceedings. The early authorities are not SEP cases, but they show how courts expect the technology to be used: to organise and analyse extensive source material while lawyers, experts and judges remain accountable for the conclusions. That model fits portfolio-wide SEP analysis remarkably well.

The UK: a useful litigation tool, subject to professional oversight

In Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank, decided on 6 June 2025, the English High Court said:

"Artificial intelligence is a powerful technology. It can be a useful tool in litigation, both civil and criminal." [7]

The Court added that AI must be used with an "appropriate degree of oversight". For legal research, professionals must verify the result against authoritative sources. The cases concerned false and unverified authorities. AI use was admitted in Al-Haroun, but the Court could not determine whether AI had been used in Ayinde. The core problem was the lack of verification and professional responsibility. [7]

Australia: courts are testing concrete AI workflows

In a speech delivered on 22 August 2026, Justice Lee of the Federal Court of Australia proposed that, once factual premises were fixed, appropriately designed AI might assemble and synthesise specialised knowledge. The parties would remain able to identify errors, challenge sources and test reliability. His conclusion was direct: "The essential point is that the Court remains responsible for the judgment." [8]

This was a judicial speech, not a ruling. But Lee J applied the same distinction in two case-management decisions only days later.

In Rogers v McDonald's Australia Ltd (AI-use), [2026] FCA 1264, the Court directed the parties to consult a technology expert and propose a protocol for possible AI use in preparing a large class action. Lee J said the process must "preserve traceability to the primary records, permit its outputs to be tested and corrected ... and remain subject to appropriate human supervision." [9]

Two days later, in R&B Investments v Blue Sky, [2026] FCA 1265, Lee J directed a similar investigation. He explained: "The point is not to allow a machine to determine what happened." Technology could organise primary information while lawyers and judges retained judgment. [10]

Both decisions concern case management. They require the parties to investigate a possible workflow. The related Federal Court Practice Note also says that an expert report should contain the expert's own opinion and reasoning. Certain uses of generative AI in preparing evidence must be disclosed. [11]

These materials do not yet decide how an AI-supported SEP report should be admitted or weighed. Their importance is directional. Courts are no longer treating AI only as a source of potential error. They are beginning to define workable conditions for its use in complex proceedings:

• every material proposition should link back to the patent, standard, prior art or product evidence;
• the methodology and important assumptions should be stated;
• errors must be capable of detection and correction;
• confidential and privileged information must be protected;
• difficult conclusions should be reviewable by a responsible professional; and
• the lawyer, expert and ultimately the court must retain their own judgment.

This is directly relevant to SEP work. A portfolio is a large body of primary material. Two different tasks must be kept separate. The first is to identify which patents could concern the functions used by the relevant products. The second is to analyse the remaining patents at claim level for essentiality, non-infringement and validity. AI can support both tasks, but the second requires far more analysis. Lawyers and experts can then focus on the issues that require judgment: claim construction, difficult validity questions, squeeze arguments and the strategic use of the results.

The emerging model is therefore AI at portfolio scale and human judgment at the decisive points.

5. From the legal challenge to a scalable evidence workflow

The legal developments above create a practical problem. Munich expects an early and compact position. Samsung v ZTE shows why broad claim-level support would be valuable. The emerging AI cases point to the controls that make large-scale work defensible. The next question is how to turn those elements into a workable portfolio process.

Consider a payment-terminal implementer approached for a 4G portfolio licence — a scenario that has become concrete since Sisvel launched its POS patent pool. Its range may include many models, regions and third-party radio modules. LTE categories and activated functions may differ. Some patents may concern network-side functions, another release or a capability the relevant configuration cannot perform.

Calling every product simply "4G-enabled" is too broad. It does not show which parts of the asserted portfolio can concern the products under negotiation.

4GLF: a product-relevance filter

ClaimsEvidence uses a workflow we call 4GLF (4G Landscaping Filter) as an early product-relevance filter. It is an internal methodology label, not a legal test. Its purpose is to identify and exclude patents that cannot concern the 4G functions used by the relevant products and LTE modules. It has three steps:

1. Map the product landscape. Record the model, SKU, region, LTE module, LTE category and supported or activated 4G functions for each relevant configuration.
2. Map the asserted portfolio at feature level. Identify the asserted claim, user-equipment or network relevance, standard specification and release, claimed LTE function, territory and legal status.
3. Match both maps. Exclude a patent as not product-relevant only where the asserted claim requires a 4G function that the documented product and module configuration cannot use or implement. LTE categories alone are not a complete functional hierarchy, so the underlying product and module evidence must be checked.

4GLF ends at that point. It does not determine whether the patents that remain are essential, infringed or valid.

A separate downstream stage: analysing the residual portfolio

The residual set then enters a separate and much more computationally intensive workflow. Each remaining patent must be examined at claim level for essentiality, non-infringement and, where relevant, validity. Material conclusions should link to the claim language, the relevant standard passage, product evidence and prior art.

The scale remains substantial even after a successful product filter. Take an illustrative pool assertion covering 20,000 patents against payment terminals. If 4GLF excludes 50% because the relevant products do not use the claimed 4G functions, about 10,000 patents remain. Those 10,000 patents still require downstream claim-level analysis. The filter makes the task more relevant and efficient. It does not make the hard analysis disappear.

This two-stage model is important. Product landscaping answers which patents could matter to these products. The downstream analysis asks whether the remaining patents actually support the licensor's asserted position.

At ClaimsEvidence, we are already combining these two layers in complete-portfolio projects. 4GLF narrows the product-relevant universe. A separate AI-native workflow then performs source-linked essentiality, non-infringement and validity analysis across the residual set. Where the evidence supports a lower share of the relevant standard stack, the resulting reports can provide the technical foundation for a lower counteroffer and licence price.

The important words are where the evidence supports it. The purpose is not to manufacture a discount. It is to test the assumptions behind the rate. This does not replace comparable-licence analysis under the Munich Guidelines. Nor may the licensor's numerator be narrowed while an inconsistent denominator is retained. The patent population, standard scope and treatment of applications, expired rights and patent families must be explained and applied consistently.

We describe this work as litigation-grade because it is designed from the outset to withstand challenge. The methodology and assumptions are stated. Material conclusions link to primary sources. Uncertainty is identified. Independent checks are built into the workflow. Difficult claim-construction, validity or squeeze issues can be escalated for expert review.

The result supports two different outputs:

• a compact position for the 25-page FRAND brief, showing the method, material findings and possible rate consequence; and
• an auditable evidence base from which the most relevant reports, patents, counterexamples and primary documents can be selected and defended.

The aim is not to file 10,000 reports. Portfolio-wide analysis is used to identify and quantify the decisive points. The court-facing material can then remain focused.

6. What an implementer should prepare for FRAND FIRST

The legal test and the technical analysis cannot be prepared in separate silos. A product filter alone does not justify a royalty adjustment. A claim-level report alone does not establish willingness. The evidence has to connect the products, portfolio and rate calculation to the implementer's conduct.

An implementer considering an early FRAND hearing should, in my view, have at least six connected workstreams ready:

• A documented product scope: the products, modules, 4G functions, standards, releases, territories and periods that are actually relevant.
• A transparent product-relevance filter: the 4GLF mapping and the evidence supporting each category of exclusion.
• A downstream technical analysis: source-linked essentiality, non-infringement and validity work across the residual portfolio, with focused human review of the difficult points.
• A comparable-licence position: why the agreements relied upon are or are not economically comparable, including product, portfolio, volume, duration and past-use differences.
• A consistent top-down cross-check: the aggregate royalty burden, relevant patent population, licensor's numerator and a methodology applied consistently to both sides of the fraction.
• A willingness record: the counteroffer and its calculation, the negotiation history, payment of the undisputed amount and any required security.

The 25-page submission should present the decisive results, not recreate the full project. Its credibility will depend on the deeper record behind it. No single report establishes willingness. No payment proves that a low figure is correct. The strength lies in the connection between conduct, calculation and evidence.

7. Conclusion: a possible synergy, not a guaranteed outcome

The Munich Guidelines probably did not have AI in mind as their direct catalyst. They are a procedural and economic proposal from one chamber. They have also attracted criticism, particularly from the implementer side. Concerns include the early payment burden, reliance on licences selected by the SEP holder and the risk that the corridor analysis may increase injunction pressure. [13]

It would therefore be premature to say that FRAND FIRST will necessarily improve SEP licensing. Much will depend on how the Chamber treats well-supported lower counteroffers, challenges to the comparability of licences and technical evidence that changes the portfolio inputs.

There is nevertheless a promising potential synergy. FRAND FIRST asks the parties to identify the decisive issues early and explain them briefly. AI can make the much broader work behind that brief possible. It can help both sides test portfolio assertions, find weak assumptions and bring source-linked evidence into the negotiation before positions become entrenched.

For SEP owners, that may mean showing portfolio strength beyond a short proud list. For implementers, it may mean demonstrating why certain patents are not relevant to the products and why the remaining portfolio supports a lower rate. Better information does not guarantee agreement. It can, however, narrow the real dispute and make a lower counteroffer more than a negotiating number.

My personal prediction is that AI will change the expected standard of preparation before it changes the formal law of evidence. Within the next one to two years, strong SEP positions will increasingly combine commercial comparables with product-specific, source-linked portfolio analysis. Sampling will remain useful. But unexplained samples and bare declaration counts will become harder to defend when broader testing is available.

This development will not automatically favour implementers or SEP owners. It should favour the party that can explain its number, connect it to primary evidence and withstand challenge.

If the Munich approach is applied openly and permits strong evidence to affect the relevant inputs, AI may become an enabler of a more efficient and better-informed FRAND process. If the corridor is applied too rigidly, the same early procedure could instead front-load pressure on implementers. That is why the first cases applying the Guidelines will matter so much.

I would be interested to hear whether SEP owners and implementers expect this combination to narrow disputes, or whether it will simply move the dispute over valuation assumptions to an earlier stage.

Sources reviewed

[1] LG München I, 7th Civil Chamber, FRAND Guidelines, dated 13 August 2026. English convenience translation made available by Dr Stephan Dorn (CMS) via LinkedIn: https://www.linkedin.com/posts/stephandorn_7th-civil-chamber-munich-i-court-frand-guidelines-activity-7498430875571986432-2Jm0/

[2] Boehmert & Boehmert, "New FRAND Guidelines of the 7th Civil Chamber of the Munich I Regional Court: More Structure, More Economics, No Safe Harbor", 28 August 2026: https://www.boehmert.de/en/new-frand-guidelines-of-the-7th-civil-chamber-of-the-munich-1-regional-court/ See also Grünecker, "Munich District Court Publishes Comprehensive FRAND Guidelines for SEP Litigation", 31 August 2026: https://grunecker.de/en/insights/munich-district-court-publishes-comprehensive-frand-guidelines-for-sep-litigation/

[3] LG München I, judgment of 8 January 2026, case 7 O 5007/25, particularly headnotes 1–3 and [97]–[115], [164]–[176] (first-instance judgment; appeal listed as OLG München 6 U 573/26 e): https://www.gesetze-bayern.de/Content/Document/Y-300-Z-GRURRS-B-2026-N-791?hl=true. See also press release no. 3 of 9 February 2026: https://www.justiz.bayern.de/gerichte-und-behoerden/landgericht/muenchen-1/presse/2026/3.php

[4] LG München I, judgment of 5 February 2026, case 7 O 7655/25, particularly [127]–[139], [153]–[162], [176]–[182] and [193]–[194] (first-instance judgment): https://www.gesetze-bayern.de/Content/Document/Y-300-Z-GRURRS-B-2026-N-2292?hl=true

[5] LG München I, judgment of 30 April 2026, case 7 O 64/25, particularly [137]–[140] and [161]–[162] (first-instance judgment): https://www.gesetze-bayern.de/Content/Document/Y-300-Z-BECKRS-B-2026-N-8886?hl=true

[6] Samsung Electronics Co Ltd v ZTE Corporation, [2026] EWHC 999 (Pat), 1 May 2026, particularly [137] and [330]–[337]: https://www.judiciary.uk/wp-content/uploads/2026/05/Samsung-v-ZTE-FRAND-judgment-REDACTED-Final-for-hand-down.pdf

[7] Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank, [2025] EWHC 1383 (Admin), 6 June 2025, particularly [4]–[8]: https://www.judiciary.uk/wp-content/uploads/2025/06/Ayinde-v-London-Borough-of-Haringey-and-Al-Haroun-v-Qatar-National-Bank.pdf

[8] Justice Michael Lee, "All the Right Notes: Artificial Intelligence and the Future of the Common Law", Sixteenth Sir Harry Gibbs Memorial Oration, 22 August 2026, section I. This is a speech, not a judgment: https://www.fedcourt.gov.au/digital-law-library/judges-speeches/justice-lee/lee-j-20260822

[9] Rogers v McDonald's Australia Ltd (AI-use), [2026] FCA 1264, orders made 25 August 2026, reasons published 28 August 2026, Order 7 and [13]–[16]: https://www.judgments.fedcourt.gov.au/judgments/Judgments/fca/single/2026/2026fca1264

[10] R&B Investments Pty Ltd (Trustee) v Blue Sky Alternative Investments Ltd (in liq) (Security for Costs and AI), [2026] FCA 1265, orders made 27 August 2026, reasons dated 1 September 2026, Order 6 and [18]–[28]: https://www.judgments.fedcourt.gov.au/judgments/Judgments/fca/single/2026/2026fca1265

[11] Federal Court of Australia, Use of Generative Artificial Intelligence Practice Note (GPN-AI), 16 April 2026, particularly [4.9]–[4.12]: https://www.fedcourt.gov.au/law-and-practice/practice-documents/practice-notes/gpn-ai

[12] Bundesgerichtshof, judgment of 27 January 2026, case KZR 10/25, FRAND-Einwand III, particularly [63], [89]–[92] and [96]: https://www.bundesgerichtshof.de/SharedDocs/Entscheidungen/DE/UebrigeSenate/KartS/2025/KZR__10-25.pdf?__blob=publicationFile&v=1

[13] Fair Standards Alliance, "The Fair Standards Alliance raises concerns regarding the purported 'FRAND Guidelines' published by the 7th Civil Chamber of the Munich I Regional Court", 24 August 2026: https://fair-standards.org/news/the-fair-standards-alliance-raises-concerns-regarding-the-purported-frand-guidelines-published-by-the-7th-civil-chamber-of-the-munich-i-regional-court/

[14] LG München I, judgment of 22 January 2026, case 7 O 4102/25, particularly [117] on supplementary security: https://www.gesetze-bayern.de/Content/Document/Y-300-Z-GRURRS-B-2026-N-1112

Note: This analysis is based on public sources reviewed on 8 September 2026. It reflects the author's personal view and is not legal advice.