Unverified50% confidenceEventExact time
In 2024, Johann Rehberger demonstrated an indirect injection attack targeting ChatGPT's memory feature, causing the model to persistently execute malicious behavior through instructions embedded in user-uploaded documents.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedChatGPT Memory and Claude Projects represent the common evolutionary direction of persistent memory in current mainstream AI products.69% similarUnverifiedAnthropic's Claude and OpenAI's ChatGPT have successively introduced persistent memory features.66% similarUnverifiedMainstream AI products such as ChatGPT Memory, Claude Projects, and Notion AI share a common direction of achieving user-level personalized adaptation through external storage and dynamic context injection.63% similarUnverifiedThe Flush+Reload attack exploits the physical sharing characteristics of shared memory pages and the CPU last-level cache, inferring the victim's memory access patterns with nanosecond-level precision by measuring the time difference for cache line reloading.61% similarUnverifiedChat Memory simulates memory by injecting historical messages into the Prompt with each request, causing token consumption to grow linearly with the number of conversation turns.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/892856API
curl https://kongchang.com/api/v1/knowledge/claims/892856MCP
get_claim(id=892856)