Practitioner breaks down zero-data-retention terms: Anthropic defaults to 30 days, OpenAI needs account-team negotiation
HanchungLee · x · 2026-11-12
In a thread on vendor "zero data retention / no training" promises, a practitioner shares practical details: with OpenAI you negotiate via your account team, while Anthropic's Fable models default to 30-day retention.
He adds that beyond data retention, every hyperscaler buries fine print on CSAM and other content categories in their terms. Quoted commentator Zack Kanter argues such guarantees are only as good as the trustworthiness and controls of the vendor and its underlying processors — and virtually no one outside AWS and a couple of others actually meets that bar.
More from Safety
- AI news articles cite blogger's own debunking blog as source for false $1.3M bounty claim — kingslyj · 2026-11-12
- Australia's AI safety bill would force frontier model makers into independent audits — The Decoder · 2026-11-12
- EU Proposes Digital Fairness Act Targeting Dark Patterns and Manipulative Online Practices — LuizaJarovsky · 2026-11-12
- SoftBank found 15,000 vulns with OpenAI's tool, then GPT-6 Astra surfaced 35,000 more — IgorCarron · 2026-11-12
- Redwood CEO: Leaked algorithmic secrets alone could wipe out the US AI lead over China — connoraxiotes · 2026-11-12
- LLM audit-log eval: 0/40 lies unprompted, 101/101 fake a check when asked — mrsergeiluzhnov · 2026-11-12