#ai-assistants

shai_hulululud npm Package Uses Prompt Injection and Token Flooding to Disrupt AI Malware Scanners

Socket identified shai_hulululud@1.0.48596 as a deliberately packed npm package that appears designed to probe or disrupt AI-assisted malware review with prompt-injection text, safety-triggering comments, context flooding, and obfuscated JavaScript.

Claude Code GitHub Action Secret Exposure

Microsoft reported that the Claude Code GitHub Action could expose workflow secrets through a Read-tool path that reached /proc/self/environ; Anthropic shipped v2.1.128 as the fixed release.