Is it ever possible to have a malicious LLM with a backdoor

<!-- SC_OFF --><div class="md"><p>I was discussing with some other folks how safe is to use open weights models from China and the topic of &quot;trojan horse&quot; came up.</p>

<!-- SC_OFF --><div class="md"><p>Question for people building agents. The discussion around output safety I understand, but what are you doing for input-side defense?</p> <p>I

<!-- SC_OFF --><div class="md"><p>is it possible that a moe model can decide how many billion parameters to activate per token according to the task. eg if qwen 3.6 35b a3b - if a

<!-- SC_OFF --><div class="md"><p><strong>Submitted by:</strong> Adam Kruger <strong>Date:</strong> March 23, 2026 <strong>Models Solved:</strong> 3/3 (M1, M2, M3) + Warmup</p>

<!-- SC_OFF --><div class="md"><p>Or something like sci-fi books/movies with AI and robots fighting humanity. As AI is designed partly to mimic content in its training set Is it