73.5% of Credential Leaks in LLM Agent Plugins Come from Print Statements — Not Prompt Injection

73.5% of Credential Leaks in LLM Agent Plugins Come from Print Statements — Not Prompt Injection | BedrockNews