The case for halting frontier AI development is gaining political traction, which is bad news for the handful of companies spending hundreds of billions to build systems smarter than people.

• AI is already dangerous in narrow ways: models now beat humans at finding software flaws, and one hallucinated bad intelligence that nearly triggered a US strike on a Chinese ship.
• The real goal of the top labs is AI that improves itself, which means handing over control on purpose and could jump from human-level to far beyond in weeks.
• Journalist Garrison Lovely argues the "China will beat us" excuse is weak, since Beijing is the last government on earth that would let machines run without supervision.
• Mass job loss is the point, not a side effect — it is the old dream of getting labor without workers, with no unions and no HR.
• A Bernie Sanders–backed bill would treat frontier model training like nuclear weapons work, with embedded auditors and prison time for violations.

Outlook: Only three companies have ever built a frontier model, so a US-China deal to pause could be far easier than people assume — but lawmakers have shown little appetite to slow anything down.