SignalNov 22, 20261 min read3 sources

Prompts sent to model providers stay stored after the request ends

In one minute

Together AI says an unauthorized party downloaded stored samples of prompts sent to its serverless inference service in June and July, using a credential exposed in a public code repository.

Dragos found that tracking links planted in AI requests were opened again days later. Google says Gemini keeps chats for about two months even with History off. Builders should treat anything sent to a provider as stored data with its own exposure risk.

From the Field Notes:

Evidence / sources

  1. 1together.ai/blog/security-event-prompt-leakage-for-requests-in-june-and-july-of-2026
  2. 2tedt.org/The-AI-Finished-its-Job-Three-Days-Later-Something-Re-opened-the-Link
  3. 3support.google.com/gemini/answer/13594961?hl=en
All Signals

Next reads

Read next

Open the graph
  1. 01topicPromptsExplore the prompts topic hub.
  2. 02archiveSignalsOpen the signals and the raw-signal archive.
  3. 03sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract