Still, any AI model vulnerable to cogitohazards is a huge risk because any model could trivially access the full corpus of human knowledge. It makes more sense

Still, any AI model vulnerable to cogitohazards is a huge risk because any model could trivially access the full corpus of human knowledge. It makes more sense to making sure the most powerful models are resistant to cogitohazards rather than developing elaborate schemes to shield their vision and hope that plan works out in perpetuity.