Study: 96% of ChatGPT memories are created unilaterally by the system
alex_verem · x · 2026-10-22
Sharing an arXiv paper: via GDPR data requests, researchers obtained ChatGPT histories from 80 users containing 2,050 memories — 96% created unilaterally by the system, 28% containing GDPR personal data and 52% psychological inferences. The paper also introduces Attribution Shield, a tool that rewrites queries to reduce personal detail exposure. See the author's full thread for details.
More from Safety
- OpenAI redacted 23 of 61 pages in METR's embedded misalignment-monitor report, researchers say — peterwildeford · 2026-10-23
- After eMudhra fiasco, AI agents set to audit certificates from 23 Indian CAs — kingslyj · 2026-10-23
- Security legend Halvar Flake: Rejecting AI-assisted bug reports over their origin is out of place — mboehme_ · 2026-10-23
- 8B safety classifier trained on 1,275 prompts beats GPT-5.4 across 5 benchmarks — HaggaiMaron · 2026-10-23
- DeepMind's Geoffrey Irving: alignment may need cryptography-grade rigor as AI finds 12 OpenSSL zero-days — geoffreyirving · 2026-10-23
- DeepMind's Irving calls to halt AI progress now, with treaties to follow — davidmanheim · 2026-10-23