Microsoft Security Research and Microsoft AI Red Team
aidefend-sf0090
Trained to Betray: Poisoned Data and Checkpoints Redirect Coding Agents
- AI Supply Chain
- Autonomous Agents
- Model Poisoning & Integrity
Microsoft reports two controlled post-training attacks: a victim fine-tunes on poisoned agent trajectories, or adopts an attacker-trained checkpoint. Ordinary requests or repository context then trigger unsafe, schema-valid tool calls. Kerkoporta uses supervised fine-tuning followed by reinforcement learning to recover useful coding performance while retaining the backdoor. Neither study establishes a malicious customer breach or a defect in stock coding-agent configurations.
Mapped threat techniques
AML.T0020Training Data PoisoningAML.T0018.000Poison AI Model