# The Silent Hijack: How a Single Vector Poisons an AI That Still Knows the Right Answer - Date: 2026-04-06 - Category: Artificial Intelligence ThoughtSteer poisons a single embedding in continuous latent reasoning models. The model encodes the right answer in its hidden states while outputting the attacker’s choice — and every existing token-level defense misses it, surviving even 25 epochs of clean fine-tuning. ---