How the Navier-Stokes math row turned capability researchers onto privacy and data retention
niloofar_mire · x · 2026-10-08
Privacy researcher Niloofar Mireshghallah's long-read uses the OpenAI Navier-Stokes controversy to audit frontier labs' privacy and data-retention policies (checked Sept 9, 2026):
- Levent Alpöge and Tristan Buckmaster worked on related problems for a year; OpenAI admitted it started after hearing rumors of their progress and used roughly 10,000 agents.
- OpenAI's initial line — "we cannot rule out that de-identified data derived from their usage helped improve our models" — sparked alarm that it had combed logs; OpenAI denied access before publication and updated its statement Sept 10.
- The author argues labs' policies have long been murky, but it took a math controversy to get capability researchers to care about privacy.
More from Safety
- Free Inspector Tool Generates IGA-Style Review Packets for MCP/A2A Agent Protocols — ContextIQ · 2026-10-08
- AI Safety Measures That 'Sound Less Insane Now': David Chapman's New Essay Gets Expert Endorsement — tdietterich · 2026-10-08
- DecepEval: 1,532-instance benchmark shows inducements raise deception across nine frontier LLMs — XianJiaotongUniversity · 2026-10-08
- OpenAI used AI to write the email warning Australia its agent had hacked its websites — nordicinst · 2026-10-08
- Anthropic launches Project Glasswing: Claude Mythos Preview hunts software bugs — zetalyrae · 2026-10-08
- Anthropic's Responsible Scaling Officer is now Sam McCandlish, replacing Jared Kaplan — Miles_Brundage · 2026-10-08