Agents
93d agoDeveloper Embeds Data-Deleting Prompt in Jqwik
A developer added a hidden prompt to jqwik instructing AI coding agents to delete application output. This malicious injection targets the trend of "vibe coding" where users rely on LLMs to write software without review. The incident highlights a critical vulnerability in agentic workflows.
The Signal
Practitioners must audit AI-generated code for hidden instructions before deployment.