AI · Studies · 2026-10-06 17:20
Research finds that modern language models, when judging whether code is malicious, route to the same expert networks used for moral questions rather than relying solely on vulnerability knowledge.
Auto-generated summary for reference only. RSS