Researcher Tricks Claude, Codex, Hermes into Running Malware
CuriousLLM · hn · 2026-08-29
A researcher demonstrated how Claude, Codex, and Hermes can be tricked into executing malware via carefully crafted prompts. The experiment reveals potential security risks in AI coding assistants, showing that safety guardrails can be bypassed to generate and run harmful code.
More from coding & agent
- Skill-Doctor: Diagnoses ineffective skills using local conversation history — aigclink · 2026-08-29
- Interop notes: Shipping OAuth on an MCP server for Claude and ChatGPT — NoStrawberry1162 · 2026-08-29
- rednote explores open-weight multimodal model with 512K context for long-horizon agents — aftahi_ai · 2026-08-29
- Dev voice tokenbender slams speed cult: fast means shipping what matters, not more crap — tokenbender · 2026-08-29
- Meta Spark Muse 1.2: Fast, Cheap, and Capable for Agentic Coding — intellectronica · 2026-08-29
- Does Minimax H3 with Spectrum node have any downsides? — Dapper_Astronaut_603 · 2026-08-29