Search
The founder of OpenClaw advises users not to use small models for high-risk tasks due to very weak prompt injection protection
20 hour ago /
Read about 0 minute
Author：小编
Recently, a netizen shared a screenshot on the social media platform X showing the GPT-5.4 model configured in their OpenClaw AI assistant, mentioning that GPT-5.4 is slower than Claude Haiku-4.5 (a lightweight small model). This post caught the attention of OpenClaw founder Peter Steinberger, who reminded users not to use small/old models like Haiku, which lack prompt injection protection, for high-risk tasks, as they are easily susceptible to manipulation through injection and are only suitable for simple tasks or permissionless environments.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
