Unverified50% confidenceFactExact time
In large language model services, the consequences of ATO take on new particularities: attackers can read historical conversations (which may contain code, documents, trade secrets, personal privacy), can call the model API as the victim to consume quota or trigger charges, and can even exploit the model's "memory" feature to poison the context.
1
Sources
50%
Confidence
Long-term
Relevance
10/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter an account is maliciously taken over, the attacker can read historical conversations and may initiate new interactions as the user to conduct social engineering attacks or abuse the service.70% similarVerifiedIndirect injection refers to attackers hiding malicious instructions in external content that the model will read, such as webpage body text, email body text, or document content.67% similarUnverified提示注入在Agent系统中尤为危险，因为被攻陷的Agent可能拥有发送邮件、删除文件、调用支付API等真实世界权限66% similarUnverifiedMemory poisoning is an evolved form of prompt injection attack, in which an attacker induces an AI to write false information into long-term memory storage, thereby influencing the system's subsequent interactions with all users.66% similarUnverifiedEntering passwords into an AI chat box is an extremely dangerous behavior, because user input is typically transmitted and stored as conversation history on the service provider's servers, posing a risk of exposure.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1119944API
curl https://kongchang.com/api/v1/knowledge/claims/1119944MCP
get_claim(id=1119944)