AI – it came, it saw, it conquered.
If 2024 was the year of AI experimentation and 2025 was the year of implementation, 2026 is the year of accountability.
For MSPs, the stakes have never been higher. You are no longer just "reselling" or "implementing"; you are often deploying sophisticated AI wrappers, customised Copilot extensions or proprietary LLM-driven services.
The problem? Most standard Master Service Agreements (MSAs) are still stuck in 2022. They treat software as a static entity and data as something that just sits in a database. AI doesn't work like that. It’s fluid, it’s ever-evolving and it’s hungry for data.
If you haven't updated your terms to reflect the reality of Generative AI, you are likely sitting on a ticking time bomb of IP ambiguity and uncapped liability.
Here are the 6 essential provisions you need to be negotiating right now to protect your MSP.
1. Future-Proofing by defining the ‘Output’
In the old world, ‘Deliverables’ were easy to define in the MSA or Statement of Work: a migration report, a piece of custom code or a configured tenant. Something fixed. In the AI world, the ‘Output’ could be anything from a simple dataset to a dynamically generated UI. Your contract needs to define Output broadly enough to capture everything the AI produces, while clearly separating it from your intellectual property that created it.
2. Build a contractual Fortress around your AI
This is where most MSPs get caught out. If your contract says "all IP created during the services belongs to the Client," you might accidentally be handing over the rights to your proprietary prompts, fine-tuning datasets or the specific configurations that make your AI solution valuable. You need to build a ‘Fortress’ with words; a clause that explicitly carves out your pre-existing AI tools and any improvements made to them during the project.
3. Obligate human oversight
AI hallucinations are a reality, not a bug. If your AI tool provides a false answer, suggests a piece of code that has a security vulnerability or a marketing summary that is accidentally defamatory - who is responsible? Your contract should mandate that the client is the final judge. By requiring a "Human in the Loop" to review and approve all AI Output before it is used in a live environment, you shift the operational risk back to the user.
4. The anti-training shield
Your IP is your competitive advantage. You must ensure that your clients cannot take the outputs or the access you provide and use them to train their own competing models. Without an explicit anti-training or anti-reverse engineering clause, you are essentially teaching your clients how to replace you.
5. Your training rights
On the flip side of the previous provision, you need the right to get smarter. To improve your services, you should negotiate the right to use anonymised, de-identified data from the engagement to train and tune your own internal models. This is the engine of growth for an AI-first business, but it must be handled so as to remain GDPR compliant.
6. The Hallucination Disclaimer
Standard warranties usually promise that services will be "performed with reasonable skill and care." AI complicates this. An AI can be perfectly designed and still produce a hallucination (a confident, but false statement). You need a specific disclaimer that clarifies that because of the nature of AI, you cannot guarantee 100% accuracy (or even fitness for a particular purpose) for every single output.
Is your MSA AI-ready?
At Law 365, we’ve spent the last year deep-diving into the intersection of tech law and Generative AI to ensure our clients aren't just protected, but are positioned to lead.
Don't wait for a dispute to find out your contract isn’t fit for purpose!
Contact Law 365 to prepare your contracts for an era of AI.
