In the LegalTech ecosystem, the race to build the “ultimate legal copilot” is accelerating. Platforms that promise to draft pleadings, summarise case law and answer legal questions in seconds all need the same thing to work: data, lots of data. And that’s where the Fastcase vs Alexi case becomes a very clear warning for any company that is training AI models with third-party databases.
In this article, we analyse what is happening in this dispute, what lies behind it from the perspective of data licences, trade secrets and trademarks, and what founders, executives and in-house counsel working with generative AI should already be reviewing.
1. What is happening in the Fastcase vs Alexi case?
Fastcase is a well-known legal research company that, after merging with vLex and later being acquired by Clio, is now part of one of the largest LegalTech ecosystems in the world.
Alexi is another LegalTech company that offers an AI-based platform to help lawyers obtain automated legal answers and analysis. For years, the two companies had a commercial relationship: Alexi held a licence to access Fastcase’s online library.
According to the lawsuit filed in Washington, D.C., Fastcase accuses Alexi of having gone far beyond that licence by:
- Training its AI models with data from Fastcase’s library.
- Displaying case law and content originating from Fastcase within its own platform.
- Using Fastcase’s trademarks and distinctive signs in a way that allegedly created the impression of an ongoing affiliation or partnership.
For Fastcase, this is not just a commercial disagreement: it is an issue of breach of contract, misappropriation of trade secrets and trademark misuse. Among other remedies, the claim seeks monetary damages, an order to stop using its data, and the destruction of datasets and models trained with that information.
Alexi, for its part, denies the allegations and maintains that it acted within the scope of the licence, describing the lawsuit as an attempt by Clio to rewrite the agreement once it perceived Alexi as a direct competitor.
2. What’s at stake: licences, secrets and trademarks in legal AI
Although this is a dispute between two players in the LegalTech world, the underlying issues affect any company that uses third-party data to train AI.
2.1 Data licences and limits on model training
The licence binding Fastcase and Alexi, according to Reuters, prohibited publishing or distributing parts of Fastcase’s database. The core of the conflict is whether training an AI model with that data and then showing results based on it falls within that prohibition.
Here, several key questions intersect:
- Is model training considered a form of “reproduction” or “use” that is relevant for contractual and copyright purposes?
- Is the output generated by the AI “new content”, or a re-use of the original database?
- What happens if the model memorises or reproduces substantial parts of the licensed information?
Other recent cases, such as Advance Local Media v. Cohere, already show that courts are beginning to accept that even AI-generated “summaries” can infringe copyright if they reproduce protected expression, not just facts.
2.2 Trade secrets in legal databases
Fastcase also alleges that Alexi took advantage of elements protected as trade secrets, for example:
- The way legal information is organised, classified and enriched.
- Proprietary taxonomies or annotations that add value compared to public sources.
For many companies, competitive advantage lies not only in raw data, but in how that data is curated, labelled and structured. If those elements are used to train someone else’s model, the debate is no longer just about copyright, but also about protecting know-how and trade secrets.
2.3 Trademarks and appearance of affiliation
The third front is trademarks: Fastcase claims that Alexi used its marks and source identifiers within the interface, creating the impression that there was an ongoing alliance or that the content was still being served by Fastcase under its control.
In an environment where legal copilots and automated assistants are proliferating, clarity about who is responsible for the content is critical. Careless use of trademarks can lead to:
- Claims for trademark infringement.
- Allegations of unfair competition or misleading advertising.
- Reputational damage if the AI-generated content is incorrect, outdated or biased.
3. Key risks for founders and legal teams
Beyond who is right in this specific lawsuit, the case offers several lessons for founders, executives and in-house counsel who want to sleep at night with their generative AI projects.
Your data provider today can be your competitor tomorrow.
The line between “customer”, “provider” and “competitor” blurs when everyone is building their own models. That makes licence and data-use clauses strategic, not just a technical annex.
Models and their weights can become both the asset and the problem.
If a court concludes that a model is contaminated by data used outside the licence, it is not only the dataset that is at risk; the model’s weights, its ability to be commercially exploited and even future funding rounds may also be in jeopardy.
The risk is not just copyright.
Current litigation shows a cocktail of legal theories: contract, copyright, trade secrets, trademarks, unfair competition, data protection… A well-crafted claim can attack the business from several angles at once.
Europe is serious about transparency and copyright.
Rulings such as GEMA vs OpenAI in Munich and the EU’s work on a Code of Practice and detailed summaries of training data point to a future in which being opaque about datasets will be very costly.
4. Practical checklist for training AI with third-party data
If your company is building a legal assistant, a contract analysis tool or any AI product trained with third-party data, now is the time to review your strategy.
4.1 Questions your licence agreement must answer
When negotiating licences for databases, content or datasets, make sure the contract clearly answers at least the following questions:
Is AI model training explicitly allowed?
It’s not enough for the contract to be silent. Ideally, it should expressly state:
- Whether training is permitted or prohibited.
- Under which conditions (purposes, territory, term).
What can be done with the model outputs?
- Is displaying fragments of the original database to users authorised?
- Is there a limit on the percentage or length of text that can appear in the outputs?
Who owns the trained models and weights?
- Does the party supplying the data have any rights over the resulting model?
- Can that model be re-used for other clients or products?
How is know-how and trade secrets protected?
- What is considered a trade secret within the database?
- Are there obligations on security, access control and technical audits?
What happens when the relationship ends?
- Must the dataset be deleted?
- Must the model be “untrained”, or are commitments not to use it in the future sufficient?
- Are there specific clauses on destruction of copies, logs and backups?
4.2 Technical and compliance best practices
Beyond contracts, there are technical measures that reduce risk:
Limit direct exposure of protected content.
Avoid having the AI reproduce decisions or articles “as is”, unless there is a licence to redistribute them.
Implement guardrails to minimise verbatim memory.
Adjust parameters and training techniques to reduce memorisation and reproduce fewer exact sentences from the source.
Dataset traceability.
Maintain a clear inventory of which sources are used, under which licences and for which models. This is essential if, in future, you are required to prove data provenance before courts or regulators.
Regular reviews with your legal team.
As the product evolves, the legal analysis should evolve as well. Using data only to train an internal prototype is not the same as launching a global legal AI platform.
5. How Brandlex can help you
At Brandlex, we support startups, scaleups and established companies that are building generative AI and LegalTech products from Spain and Chile to the world. We encounter the same dilemmas raised by the Fastcase vs Alexi case every day:
Can I train my model with this database?
What if tomorrow the provider changes strategy and decides to compete with me?
How transparent do I need to be about my datasets under the AI Act and European copyright rules?
From Brandlex we can help you:
- Design and negotiate data licences that are compatible with generative AI.
- Identify and protect key intangible assets (databases, taxonomies, models, trademarks).
- Review your IP, data and contractual strategy so that your legal copilot, analysis platform or content marketplace is not built on a legal time bomb.
If you are developing or using AI in your business and want to make sure your data licence for generative AI is robust, this is a good time to review your contracts, policies and technical architecture with a team that lives at the intersection of IP, technology and AI regulation.
👉 Would you like to review your data and AI strategy?
You can contact us to explore how to safeguard your product from Spain and Chile to the markets of Europe, Latam and Asia.
At BRANDLEX, we can help you align innovation, copyright, and regulatory compliance.
👉 Write to us at info@brand-lex.com or visit our website to learn how we can work together.
Disclaimer
This article is for information purposes only and does not constitute legal advice. Each AI project and each data licence agreement must be analysed on a case-by-case basis. To obtain specific legal advice, we recommend contacting the Brandlex team directly or another professional specialised in intellectual property, data and artificial intelligence.