Unverified50% confidenceFactExact time
When an autonomous agent is able to read and execute text instructions embedded in a webpage, any text on the page can potentially serve as a vector for prompt injection.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPrompt injection can be divided into direct injection and indirect injection, where indirect injection delivers instructions through external documents, web pages, and other content read by the model.68% similarUnverifiedTool-calling agents can not only generate text but also invoke search engines, execute code, manipulate file systems, and even initiate network requests.67% similarVerifiedPrompt injection is divided into two categories—direct injection and indirect injection—where indirect injection refers to attackers concealing malicious instructions in external content that the AI may read, such as webpage body text, code comments, README files, and image OCR text.66% similarUnverifiedIn Coze's autonomous planning mode, the description text of a workflow is the sole basis on which the AI decides whether to invoke that workflow.66% similarUnverified规则的本质是系统提示词的扩展注入,Agent启动新会话时会在加载核心系统提示词之后、接收用户输入之前自动读取配置文件中的规则并注入到对话开头65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901286API
curl https://kongchang.com/api/v1/knowledge/claims/901286MCP
get_claim(id=901286)