Boss Moves: How founders can future-proof IP licensing contracts

Expert advice to prevent today’s AI data agreement from becoming tomorrow’s liability.

In this series we tackle some of the thorniest questions facing business leaders today.

Dear Anxious,

If you have concerns about your company’s reliance on data you don’t fully control, you’re not alone. As AI-powered products mature and companies get acquired, licensing agreements are being stress-tested in real time.

One case currently unfolding in the legal tech world offers a cautionary tale. (And if companies in that sector are struggling with data agreements, none of us are immune.) In 2021, Toronto-based startup Alexi licensed U.S. case law from American legal research platform Fastcase to power its AI-generated legal memos. The partnership ran smoothly for years — until Fastcase was acquired as part of a billion-dollar-deal by Canadian legal software giant Clio. During due diligence, Clio discovered a clause that could benefit a future Alexi acquirer and, according to court filings, pressured Alexi to relinquish it. When the startup refused, Fastcase sued for breach of contract, alleging Alexi had used the data commercially rather than for internal research. Alexi denies the allegations, arguing that its data use never changed — only Fastcase’s incentives had.

The dispute underscores the new reality: What once felt like a stable agreement can become a point of contention as products scale, companies are acquired or competitive dynamics shift.

Anthony de Fazekas, a partner at Mintz specializing in IP strategy and governance and a volunteer advisor at MaRS, helps founders figure out how to develop data agreements that can withstand the test of time. De Fazekas works closely with startups and large enterprises, often stepping in when growth, financing or acquisitions force companies to take a harder look at their contracts. Here’s what he recommends.

Secure your rights to use information

One of the most common pitfalls for AI-driven businesses is a mismatch between contractual rights and how value is actually being generated. Most AI products are built from a combination of licensed data, proprietary inputs, third-party tools and human expertise, layered together over time. And most contracts tend to follow traditional licensing models that focus narrowly on access to raw data, without addressing what you actually build with that data.

Founders can find themselves in a precarious position: they may have built something commercially viable, but their contract might not be clear about whether or not they can capitalize on what they’ve created. “A company can end up in a situation where it has value, it wants to monetize what it’s built, but it doesn’t actually have the underlying legal rights to do so,” says de Fazekas. If you can’t trace where your data came from, and what rights travel with it, defending what you’ve developed may be a challenge.

Clearly define what “internal use” actually means (before it becomes a legal dispute)

In a field like life sciences, there’s a general understanding of the difference between internal or research use and commercial use. That same line is often blurrier in AI contexts, as standard definitions don’t capture how AI products work. Data may never leave a company’s platform, remaining internal in a technical sense, yet still form the basis of commercial products that generate subscriptions, user-facing features or performance gains that drive revenue. De Fazekas says companies need to incorporate language that addresses this possibility. A safe assumption is this: if your product generates value for customers or revenue for your business, it will likely be treated as commercial use, even if the data itself remains behind the scenes.

Treat your know-how like the asset it is — or risk losing legal protection

Founders who rely on open-source models or widely available tools may assume they don’t have much IP to protect. In practice, the value of an AI business often sits elsewhere: soft IP, such as proprietary workflows, data selection and curation decisions, training methods, and the human expertise that shapes how systems are built and refined.

These elements may not look like traditional IP, but they can function as powerful trade secrets — if they’re treated that way. If you don’t take steps to actively protect how your system works — for instance, by creating a trade secret register that clearly identifies, documents and restricts access to key proprietary processes — you may be dismayed to discover that the law treats it as unprotected, regardless of how central it was to your company’s success.

Lock down the fundamentals before you sign

When advising founders entering data agreements, de Fazekas emphasizes three areas that are often overlooked and costly to fix down the road. First, get precise about what you’re actually licensing. Founders tend to focus on volume, but certain parameters may be more important: data quality, consent, compliance requirements and technical standards. If your product depends on high-quality or specialized data, be explicit about those expectations in the agreement.

The second area is what de Fazekas describes as “data flow.” AI products typically involve multiple parties contributing data, tools or expertise, with outputs generated along the way. He recommends that founders trace those contributions end to end, and make sure contracts clearly support the rights needed at each stage.

Finally, align your rights with your commercialization plan. “Sometimes you know exclusive ownership doesn’t make sense,” says de Fazekas.“While it creates a bit of messiness, joint ownership of some of these assets is usually not a bad place to land, as long as the respective rights of joint owners are clearly laid out.” The key is ensuring all agreements reflect how you intend to build, deploy and monetize your product.

Revisit your IP and data agreements frequently

For many AI companies, problems arise not because a product stops working, but because underlying agreements are no longer in sync with developing needs. “A lot of companies develop an IP strategy with a theory of their value proposition in mind,” says de Fazekas. “But the value of the company is shifting all the time.” Products evolve, data sources multiply and commercialization paths shift. That misalignment can go unnoticed — until, say, a negotiation forces a closer look. At that point, investors or buyers will ask whether existing data rights actually reflect how value is being created today.

“Companies should reevaluate their IP strategy often — almost every two months,” says de Fazekas. If they don’t, founders may find themselves renegotiating key terms under pressure, accepting a reduced valuation or, in more contentious cases, facing legal disputes. That diligence can help highlight gaps early, while there’s still room to address them on your own terms.

Are you a business leader with a conundrum? Write to us at media@marsdd.com.

Photo illustration: Stephen Gregory; Images: Unsplash