Technology Contracts · 5 min
AI vendor contracts: allocating risk when the model gets it wrong
When an AI system produces a harmful output, the contract decides who carries the loss. Most AI contracts are silent where it matters.
AI contracts inherit software terms that were never written for probabilistic systems. Performance, accuracy and liability need rethinking when the product is a model, not deterministic code.
IP in inputs and outputs is a live question. Who owns the prompts, the fine-tuning data and the generated outputs should be answered explicitly, not assumed.
Liability caps designed for licences look very different when a model's error can cascade through a production process or a financial decision.
Regulatory allocation matters too. As the AI Act assigns duties to providers and deployers, the contract should mirror who bears which obligation.
We engineer AI contracts around the failure modes that actually occur, so the allocation of risk is deliberate rather than accidental.
Altri articoli
The AI Act is live: what boards must actually do now
The EU AI Act's obligations are phasing in. The companies that treat it as a governance programme, not a one-off legal review, will move fastest.
MiCA in practice: classifying your token before it classifies you
Under MiCA, the legal nature of a token determines the entire compliance path. Classification is not a formality, it is the strategy.
NIS2 puts cybersecurity on the boardroom balance sheet
NIS2 extends cybersecurity duties deep into industry and infrastructure, and holds management personally accountable for failures.
Iniziamo a parlare
Raccontaci cosa stai costruendo.
Che si tratti di IA in produzione, di una struttura di token, di una domanda di licenza o di una fabbrica che si digitalizza, ti aiutiamo a costruirla su solide basi giuridiche.