How we’re responding to The NYT’s data demands in order to protect user privacy

Original Article ↗ Hacker News Discussion ↗

Scope and purpose of the court order

  • Many commenters see the order to preserve all ChatGPT logs (including deleted and “temporary” chats) as standard US evidence-preservation practice in a copyright case: NYT wants to quantify how often verbatim or near-verbatim NYT text is generated to calculate damages.
  • Others argue this goes far beyond normal proportionality, sweeps in huge amounts of unrelated, highly personal data from uninvolved users, and sets a bad precedent for privacy-focused services.

Privacy, logging, and “legal hold”

  • Strong skepticism that OpenAI meaningfully protects privacy: users assume everything sent to a hosted API is logged indefinitely, regardless of marketing claims or toggles.
  • Several point out that a “legal hold” is just a preservation requirement; it does not legally block OpenAI from using or accessing the data for other purposes unless other policies/laws do.
  • Some say data is a “toxic asset” and the only secure option is not retaining it at all; being forced to keep it inherently increases risk.

Zero Data Retention (ZDR) and product behavior

  • Commenters note ZDR APIs exist but are hard to actually obtain; requests are allegedly ignored, leading to accusations that ZDR is more marketing than reality.
  • OpenAI’s own post says ZDR API endpoints and Enterprise are excluded from the order, but people question why privacy is a paid/approved feature rather than a universal option.
  • There is confusion and criticism around the in-app “Improve the model for everyone” toggle versus the separate privacy portal, seen by some as a dark pattern.

GDPR and non-US users

  • Debate over whether complying with the US order violates GDPR:
    • Some say GDPR has allowances for court-ordered retention and it’s only a problem if data is kept beyond the case.
    • Others cite GDPR limits on honoring third-country orders without specific agreements and argue an EU court might bar such retention for EU residents.

NYT vs OpenAI copyright dispute

  • Several think NYT’s underlying claim is strong, pointing to examples where ChatGPT allegedly regurgitates NYT text and arguing per-infringement damages justify broad discovery.
  • Others view OpenAI’s training as fair use and call NYT’s demand overbroad or abusive of US discovery rules.
  • OpenAI’s public framing of the lawsuit as “baseless” and as a privacy attack is widely characterized as spin; critics say OpenAI’s own copyright decisions created this situation.

Government and surveillance concerns

  • A long subthread debates whether US intelligence agencies likely access such data:
    • Some assert it’s almost certainly tapped and easily searchable using modern methods.
    • Others call this unfalsifiable conspiracy thinking, noting legal and technical barriers, but still concede metadata alone is highly revealing.

Sensitivity of LLM chat histories

  • Many emphasize that LLM conversations can be more revealing than browser history: people use them for emotional processing, relationship issues, work drafts, and “raw” inner thoughts, making the retention order feel especially invasive.