A lab buying data checks where it came from and writes personal data out of the contract: arrive with the chain of custody documented and personal data already removed
Anthropic's registration statement (S-1, 26 Oct 2026) says it conducts diligence on third-party data sources for quality, provenance and suitability, and in many cases puts contract terms in place to discourage personal data in the datasets it sources. Its risk factors add that inadequate documentation of training-data sources, provenance or chain of custody could create legal exposure and complicate its answers to regulators and discovery requests. The same filing shows the spend behind this work rising: its vendor costs for data quality and performance evaluation grew $165M in 2024, $165M in 2025 and $251M in the first half of 2026 on the year before.
Dated 26 Oct 2026Source checked 31 Oct 2026Quote found on its page again 31 Oct 2026Basis: strong
“We conduct diligence on third-party data sources to assess their quality, provenance, and suitability for our usage, and in many cases include contractual restrictions intended to discourage the inclusion of personal data in sourced datasets.”
- Anthropic: increase in vendor costs for data quality and performance evaluation services (year on year), H1 2026
- Anthropic: Bartz settlement charge (copyright class action over books used in training), FY2025
- Anthropic: increase in vendor costs for data quality and performance evaluation services (year on year), FY2025
- Anthropic: increase in vendor costs for data quality and performance evaluation services (year on year), FY2024
- Anthropic and data providers and licensors (none named): buyer program, datasets licensed or commissioned from third parties, with public web data and synthetic data
It is shown on its own, not counted into a figure.