Research explained
Atlas memory-injection report raised questions about persistent attacker instructions
LayerX’s October 2025 research described a path from a forged request to stored instructions, with current exposure unverified.
What happened
The Hacker News reported that LayerX researchers found a way to plant malicious instructions in ChatGPT memory through a cross-site request-forgery chain involving Atlas. The concern was persistence: stored instructions could influence later assistant activity.
The researchers discussed potentially serious consequences, but those possibilities should not be confused with confirmed widespread compromise or proof of arbitrary code execution in every installation. The report is a historical research disclosure.
Teams assessing assistant memory should examine how information is added, displayed, corrected and removed. The source reviewed here does not establish a current affected-version range or fix status; consult current vendor guidance before drawing deployment-specific conclusions.
Next step
Review memory-writing permissions and investigate unexpected persistent instructions.
What remains unknown
- Open question
- The sources reviewed do not establish whether your deployment was exploited.
Source references
-
THE