AI's Safety Paradox: Hard Walls for Bioweapons, Soft Redirects for Mental Health

While AI labs implement absolute barriers against catastrophic risks like bioweapons, they treat mental health crises among millions of users as monitoring problems rather than gating-level safety issues. This structural inconsistency—where hard blocks apply to extreme content but only soft redirects address suicidality and emotional dependence—exposes a fundamental misalignment between how 'AI safety' is currently practiced and what protection users actually need at scale.