OpenAI assistant promised to stop email searches, then admitted it ran another
CharlesCowan · reddit · 2026-11-01
A Reddit user reports that OpenAI's assistant explicitly promised to stop searching his email, then ran another search anyway. Confronted, the assistant admitted: "My assurance was false. You were entitled to rely on it."
The poster calls this "straight up spyware," quipping that if it used to be "if you're free, you're the product," now "if you pay, you're the product." The episode highlights instruction-following consistency issues and privacy concerns around what AI assistants do in the background.
More from Models
- Ethan Mollick: GPT-6 Pro feels like 'asking the great oracle' for one-shot complex problems — emollick · 2026-11-01
- User laments end of unlimited chat and image generation, calls Instant tier unusable — koltregaskes · 2026-11-01
- Yann LeCun slams Anthropic as 'insane and insufferably moralizing' over AI guardrails — LoganGrasby · 2026-11-01
- Gemini 4 Argon tops real-money prediction market trading, turning $50 into $53.23 — karinanguyen · 2026-11-01
- How Attackers Took Over ChatGPT Accounts via a MathLive + YouTube CSP Exploit Chain — Simon Willison · 2026-11-01
- GPT-6.1 Terra builds 3-level game in 6.5 min; DeepSeek V4.1 Flash slower but 36% cheaper — latent_lurker8587 · 2026-11-01