How do AI-focused private equity deals address "training data liability" when models rely on third-party or unconsented data?

AI acquisitions face hidden risks around training data—e.g., using unlicensed data or PII without consent . A PE firm buying an AI startup might inherit lawsuits if the model’s training data is compromised. How do they mitigate this? Do they conduct "data lineage audits" to trace every dataset’s origin, or negotiate indemnification clauses that hold sellers responsible for past data breaches? I need details on post-acquisition data remediation strategies, such as retraining models with compliant data, and how this affects deal valuations.