The liability in AI training data is becoming harder to ignore. AI systems are only as good as the data they are trained on. Yet behind every dataset lies a potential legal landmine. From scraped content and proprietary material to biased historical records, training data often carries hidden exposures. As AI becomes embedded into commercial applications, these risks are no longer theoretical.
Brokers and insurers have a critical role to play in helping AI startups and scaleups identify, quantify, and transfer these risks. But first, we need to unpack where the liabilities actually lie.
What Makes Training Data Legally Risky?
There are three main sources of liability when it comes to AI training data:
- Copyright Infringement: Many AI models are trained on large-scale web-scraped data that may include copyrighted content. Even if the scraping was done indirectly via third-party datasets, liability may still exist if that data is used to generate outputs.
- Bias and Discrimination: Historical data often reflects systemic biases. When used to train models, it can result in discriminatory outcomes in hiring, lending, housing, and more — opening the door to regulatory scrutiny and civil litigation.
- Privacy Violations: Personally identifiable information (PII) embedded in training data can lead to breaches of data protection laws like GDPR or CCPA, especially if consent was not obtained or the data was not anonymized properly.
Case Study: The Getty Images vs. Stability AI Lawsuit
A defining moment in this emerging risk category is the lawsuit filed by Getty Images against Stability AI in 2023. Getty alleges that Stability AI copied over 12 million images from its licensed collection to train the popular AI image generator Stable Diffusion, without consent or compensation.
The case highlights multiple exposures:
- Use of copyrighted material without licensing
- Commercial gain from derivative outputs
- Allegations of brand dilution due to distorted Getty watermarks in AI-generated images
While the case is ongoing, it sets a strong precedent: training data is not immune to copyright law, and companies deploying generative AI may be held accountable for the data provenance behind their models.
Revisiting the Risk: What This Means for AI Companies
For companies building or deploying AI models, this case illustrates a broader principle: your liability may begin long before a model goes live.
Hidden exposures may arise from:
- Pretrained models using third-party datasets with unclear provenance
- Outputs that replicate or mimic copyrighted content
- Lack of documentation around dataset sourcing and consent
- Absence of fairness testing or bias mitigation
Insurers and brokers need to go beyond standard tech E&O templates to assess how data governance, model design, and downstream applications create risk.
Where Continuum Comes In
Continuum supports clients in identifying their risk exposures and securing the right insurance coverage to protect against them. For AI and emerging tech businesses, this may include:
Tech PI including Cyber Insurance – Covers liabilities from AI system failures, IP disputes, data breaches, and third-party claims related to data use.
D&O Insurance – Protects senior management from personal liability arising from governance decisions, including those involving training data and compliance.
Legal Expenses Insurance – Helps cover defense costs in copyright, privacy, or discrimination-related claims tied to AI models and datasets.
Start-Up Business Insurance – Tailored packages that combine essential coverages to help early-stage AI firms safeguard their operations and scale confidently.
Continuum works closely with clients and specialist underwriters to secure these policies and ensure coverage aligns with the unique risks of AI development.
Closing Thoughts
Training data is no longer a back-office concern. It is a frontline risk issue that can result in real financial, legal, and reputational damage. Brokers, insurers, and AI firms must work together to create a more transparent and defensible ecosystem.
Understanding these liabilities today is the first step to protecting against them tomorrow.
Curious if your insurance program is keeping up with your AI risks?
Get in touch to explore coverage options that support your growth and reduce liability.
