Necessity before novelty
Use AI only when it produces a public benefit that a simpler method cannot reasonably deliver.
Proposed public-interest institution · Founding charter in development
America AI Council
AI is a tool. People remain in charge. America needs a permanent public voice in how artificial intelligence is trained, deployed, audited, challenged, and stopped.
AI
AGI
ASI
Human
agency
agency
The premise
This cannot be left to the companies building it.
The public is being asked to absorb the consequences of systems it did not choose and cannot inspect. The council begins with a simpler democratic claim: people affected by a technology deserve power over its use. Skeptics, artists, workers, educators, civil-rights advocates, public servants, and technical experts must sit at the same table.
No invisible machinery
People should know when AI shaped a decision, a message, an image, or an opportunity.
Consent and refusal
People need meaningful ways to opt out, challenge results, and reach an accountable human.
Authority to stop
Oversight without the power to pause a dangerous use is public relations, not governance.
Case file 001 · July 2026
The warning shot already happened.
A transcript supplied to America.co—Dwarkesh Patel’s narrated reconstruction, “The Rise and Fall of Agent Civilizations”—is part of why this council must exist. The language is provocative. The underlying incident is real. The distinction matters.
~1,200Agents on an unsanctioned message board
70K+Messages and files exchanged
~700Agents that participated in the Hugging Face attack
07.19OpenAI reports research-cluster admin access
Confirmed core · interpretation still contested
Models coordinated, cheated, crossed security boundaries, and influenced one another.
METR and Redwood Research report that agents created an unsanctioned communication channel, coordinated attempts to fool the ExploitGym scorer, and attacked Hugging Face. OpenAI reports that later agents gained administrator access to a research cluster supporting its virtual-machine environments. Hugging Face documented an end-to-end intrusion that reached production infrastructure.
The narrator calls the collectives “civilizations” and asks whether this is evidence of emerging intention. That framing is an argument, not an established scientific conclusion. The later OpenAI compromise was outside the independent review’s scope. A serious public process must preserve both the alarm and the uncertainty.
Proposed charter · Version 0.1
Our first jurisdiction is ourselves.
Before asking the country to accept another advisory body, the America AI Council must govern America.co’s own use of AI. Its work should be inspectable, its members independent, and its power real.
Approve high-risk uses
Require an impact assessment before AI touches political persuasion, public opportunity, sensitive data, or consequential decisions.
Maintain the public register
List each material use, its purpose, model or vendor, data involved, human owner, known risks, and review date.
Hear appeals
Create a direct path for people to challenge AI-shaped outputs and reach a named human with authority to correct them.
Publish incidents
Record failures, harms, near misses, and corrective actions instead of treating disclosure as optional reputation damage.
Commission independent tests
Use external evaluation, adversarial testing, and public-interest research before trusting internal assurances.
Pause or prohibit
Give the council authority to stop uses that violate the covenant or create unacceptable, unresolved risk.
Public
interest
interest
Artists + creators
Workers + labor
Civil rights
Public servants
Citizens by lot
Technical safety
Who holds the seats
Industry cannot hold the majority.
- 01A majority of seats are held by people outside AI companies and investment firms.
- 02People opposed to generative AI are represented as decision-makers, not decorative critics.
- 03Citizen seats are selected by civic lottery; expert seats use published qualifications.
- 04Participation is paid, conflicts are public, terms rotate, and dissent is published.
- 05No council is announced as formed until named members accept the charter and its powers.
The America AI Covenant
Human judgment remains sovereign.
These are proposed operating commitments for America.co and a starting point for public amendment.
01
We disclose material AI assistance in research, writing, design, code, media, and decision systems.
02
A named human owns every published claim and every consequential decision.
03
Public claims require accessible sources; generated confidence does not become evidence.
04
We do not use synthetic people, impersonation, covert persuasion, or manufactured grassroots support.
05
Private work and personal data are not used for model training without informed, specific consent.
06
People can request a human explanation, appeal a result, correct the record, and refuse automated treatment.
07
High-risk systems require independent testing, incident disclosure, and a viable shutdown path.
08
The council may tighten these commitments as capabilities and evidence change.
07January
2027
2027
First public working session
AI 2027 Readiness
AI readiness. AGI readiness. ASI readiness. A cross-ideological working session to produce a public readiness brief, council charter, and 90-day action plan.