Unverified50% confidenceCautionExact time
In adversarial input scenarios, both typed and generative judges read the content being scored, and the content may carry instructions targeting the judge, so written reasoning cannot reliably expose prompt injection.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThere is no reliable 'detect and block' solution for prompt injection, because it is difficult to judge whether a piece of text is an ordinary description or a malicious instruction, and to the model the two are identical in form; isolation and permission control are more realistic than content filtering.71% similarUnverifiedLetting the coding agent review its own diff means it brings the assumptions it made while writing the code into the review, making it hard for it to find the issues it overlooked.71% similarUnverified用户可以通过提示注入和语义改写等手段绕过内容检测，使实际拦截效果是概率性的而非绝对的67% similarUnverifiedWhen faced with ambiguous input, HowToLiveBetter proactively asks for additional information rather than forcibly generating a possibly wrong answer.66% similarUnverifiedMalicious prompts in spreadsheets can be hidden in white-colored text, hidden columns, or comments, and once read by an agent they trigger unintended behavior.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047774API
curl https://kongchang.com/api/v1/knowledge/claims/1047774MCP
get_claim(id=1047774)