If courts and regulators start forcing labs to actually delete learned representations of copyrighted or improperly sourced data, not just filter outputs, that flips the economics of every training-data deal signed to date. Buyers like OpenAI, Google and Meta have been pricing licenses as one-time or subscription inputs; machine unlearning obligations effectively attach a contingent liability to every dataset, retroactively.
Expect data licensors and rightsholders to push for indemnification clauses and post-hoc audit rights, according to IAM Media, and expect the compliance cost of "unlearning" to become its own line item sellers can charge for.
The 'machine unlearning' problem: AI's hidden licensing and compliance challenges
