●researcherChallenges the assumption that RLHF and constitutional AI methods provide robust safety guarantees — relevant to alignment and red-teaming research directions.
●policyA federal researcher's public position that AI safeguards are fundamentally unreliable strengthens the case for mandatory external controls rather than relying on model-level alignment.