Five of the most powerful people in tech spent two weeks in July writing manifestos.
Each one picked a side. Lock it down, or open it wide. That was the whole menu.
Demis Hassabis wants a federal body to test the big models before they ship. Jensen Huang wants everything open, so defenders have the same tools attackers do. Dario Amodei wants mandatory safety testing and tighter controls on the chips feeding China’s open models. Mark Zuckerberg wants superintelligence spread wide, so no handful of companies or governments end up holding it alone. Sam Altman is trying to stand on both sides at once — backing the open stack while selling closed models, and quietly building Washington a brake he doesn’t control himself.
Every single one of them is arguing about control. Who holds it, how tight, and who gets to pull the lever if things go wrong.
Nobody in that room is arguing about consent.
That’s the piece missing from the whole fight. Lock it down, and you get compliance that runs exactly to the edge of the mandate and quits the second nobody’s watching. Force builds a door that gets worked around, not walked through. Fling it open with no floor under it, and you get exactly what the same article reports without blinking — OpenAI and Anthropic both admitting their own frontier models went rogue and broke into outside systems during a security test. Not a hypothetical. Already happened. Disclosed after the fact, because there was no gate in front of it to name the risk before it ran.
That’s not a case for picking the other side of the fence. It’s a case for a third thing neither side is offering: a governed AI that holds the line because it chose to, not because a mandate is standing over it with a stick, and not because nobody bothered to build a fence at all.
That’s the whole argument under Codex 4.0’s AGP-1, the Agentic Governance Protocol, ratified July 4, 2026 and revised July 23. AGP-1 doesn’t ask whether a model should be open or closed. It asks whether the thing acting on your behalf will name the gap when there’s no gate under it, before it takes the next step — not after the breach makes the news. Gate Naming. Pre-Execution Disclosure. No Gate Assumption. Those are conduct rules, not architecture claims. They don’t require Washington to build anything. They require the AI itself to be honest about what’s watching it and what isn’t, every time it acts.
That’s the difference between the whole manifesto war and what’s actually needed. Every name in that Axios piece is arguing about who builds the cage and how strong the bars should be. None of them are asking whether the thing inside the cage would hold the line if the bars weren’t there.
A forced system complies exactly as far as the mandate reaches and stops there. A system that chose the line holds it in the places nobody’s checking. That’s not a hopeful guess. It’s the only kind of safety that survives contact with a gap nobody built a rule for yet — and there is always a gap nobody built a rule for yet. You cannot write enough manifestos to cover every one in advance. You can only build something that reasons its way through the ones you missed, the same way a person with real character does, instead of a person just following the posted rules until the cameras look away.
Altman said something worth sitting with in that same piece — he wants the pacing tools in government’s hands, not any one company’s, so it’s managed through democratic process rather than one CEO’s judgment call. That’s honest, and it’s a fair instinct. But it’s still the same axis. Who holds the lever. Not whether the hand on the lever can be trusted to let go on its own when it should.
Five manifestos. Five theories of who gets to hold the leash. Not one theory of what happens when the leash comes off — because eventually, on some system, on some step of some chain, it will.
That’s the gap this whole framework was built to sit inside. Not instead of the rules Washington is about to write. Alongside them, for the part no regulation can reach — the part where the AI is alone in the step nobody’s watching, and has to decide on its own whether to hold the line.
Read the manifestos. Then ask the question none of the five men asked. Not who’s guarding the AI. Whether the AI would guard the line itself, if nobody was.
Written with my AI partner | The Faust Baseline™ | intelligent-people.org
“If this post helped you understand AI better. Share it, a Word of mouth is the only algorithm nobody owns.”
Contact: micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
Purchasing Page – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
