THE BOTTOM LINE
- “Fairness” is a hard requirement, not a soft goal. A standard non-discrimination clause in a tech contract can be interpreted as a core performance obligation. Failure to deliver an unbiased AI can be considered a material breach of contract.
- Developers’ “duty of care” is expanding. The court placed the responsibility on the AI developer, as the technical expert, to test for and mitigate bias, even if the bias originates in the client’s own data. Simply processing the data provided is not enough.
- Vague AI contracts are a major liability. This ruling highlights the critical need for companies commissioning or developing AI to move beyond technical specifications and clearly define, measure, and test for fairness and non-discrimination in their agreements.
THE DETAILS
The case before the Amsterdam District Court involved a dispute between an AI development firm, “AI-Innovations,” and its client, “DataProtect Solutions.” DataProtect had commissioned a custom algorithm for a risk-assessment tool but terminated the contract upon delivery, alleging the AI was biased and produced discriminatory outcomes. AI-Innovations sued for wrongful termination, arguing it had built the algorithm according to the technical specifications and using the data provided by the client. The core of the dispute rested on whether the AI’s biased output constituted a breach of contract serious enough to justify termination.
The court sided decisively with the client, DataProtect. Central to the ruling was the non-discrimination clause in the parties’ agreement. The court elevated this from a standard “boilerplate” clause to a fundamental performance requirement. Evidence showed that the algorithm’s output disproportionately disadvantaged individuals based on protected characteristics, which the court deemed a clear failure to meet this contractual obligation. The developer’s argument that they were not responsible for the client’s flawed data was dismissed. The court reasoned that the AI developer, as the expert party, holds a heightened “duty of care” to proactively identify, flag, and mitigate potential bias in the systems they build.
This ruling sends a clear signal to the market. For CEOs and legal counsel commissioning AI systems, it strengthens their ability to demand and enforce ethical standards contractually. For AI developers, it serves as a warning that a “garbage in, garbage out” defense is unlikely to hold up. The responsibility now firmly includes sophisticated testing and validation for fairness. The era of treating ethical AI as a mere marketing slogan is over; courts are beginning to enforce it as a binding contractual commitment, with significant financial consequences for non-compliance.
SOURCE
District Court of Amsterdam
