Study: Pretraining Poisoning Can Plant Agentic Backdoors

A new paper shows for the first time that poisoning pretraining data can plant a remote code execution backdoor in an AI agent. Pretraining Qwen3 models at 0.6B, 1.7B and 4B and post-training them into CLI agents, researchers found 0.1% poisoning sufficed; the 4B model reached roughly 92% attack success and stayed positive even at a 0.0005% poison rate.

• 64GB no longer a dead zone for local AI agents2026-11-01
• AWS guide: train Qwen3 on Trainium with torchtitan2026-10-22
• Pulse Local Launches On-Device AI Budgeting App for Mac2026-10-14
• 3.87B MoE trained from scratch on 86.5B tokens2026-10-05
• Qwen-family LLMs becoming backbone of audio models2026-09-30