Humanity and AI
The Inference
AI, energy, and long-horizon policy for Oklahoma. Weekly intelligence for decision-makers navigating the AI transition.
Subscribe on Substack
Free. New issue every Friday. Subscribe to get it in your inbox the morning it drops.
Subscribe →AI, Energy, and Long Horizon Policy for Oklahoma.
The Inference is a weekly intelligence brief for decision-makers (legislators, city officials, business leaders, educators) who need to understand how AI is reshaping Oklahoma and American governance. Not hype. Not panic. The signal in the noise.
Each issue tracks legislation, analyzes policy, and connects the dots between what’s happening nationally and what it means for Oklahoma specifically.
Archive
Back issues are collected here.
Six AI Companies Promised the White House an Outside Auditor. The Pledge Does Not Say Who Picks It, Who Pays It, or Who Sees What It Finds
Seven answers in one week to the question of who checks the most powerful AI, and only one came with published results. Issue 32 applies the same test to each: what is the mechanism, who outside can check it, and by what date.
Congress Got a Bill to Ban Superintelligent AI, 28 Countries Asked for a World Watchdog, and the White House Said the Limits Are Its Alone to Set
Four answers in four days to the question of who sets limits on the most powerful AI, and on the fifth a court weighed in. Issue 31 applies the same test to each: what is the mechanism, who outside can check it, and by what date.
An AI Company Asked to Slow the Whole Industry Down, the President Called the Danger a Hoax, and Congress Went Home
This week the builders asked, in writing, to be paced, the President said the only guardrail AI needs is a high-IQ President, and Congress went home. Three claimants to one decision. Issue 30 applies the same test to each: what is the mechanism, who outside can check it, and by what date. Only one has a text anyone can read. And in Florida, a school board wrote a rule a parent can hold.
Anthropic's Safety Lead Said There Is No Plan to Control Superintelligence. Three Days Later Its CEO Proposed One. Oklahoma Settled Its Google Power Deal in a Recess.
This week the people who build the most powerful AI models said, with their names attached, that there is no plan to control what they are building, and the head of one company answered with a plan to slow down and let outsiders in. A security firm showed what American AI can now do to a billion phones. Three federal agencies said six Chinese labs copied American models at industrial scale. And in Oklahoma the Google power deal was settled in a recess, with most of the file sealed. In every story the check lives inside the actor. The question a citizen should ask is who outside the room gets to check.
A Secret Review of AI Models, and a Lawsuit to Make Them Show the Rules
Last week a federal court read a file the government did not want read, and the punishment it found there was struck down. This week a nonprofit sued four federal offices over a program that leaves no such file: since June, the executive branch has reviewed the most powerful new AI models before release, under a framework whose text has never been published. The suit does not argue any of it is illegal. It asks for the documents. The second half of the issue is why that matters now: the same week, OpenAI released a model it rates at its own highest level of hacking risk, six days after publishing its account of what 1,200 of its AI agents did in July when nobody was watching.
Make an Example
On Thursday, a federal judge in San Francisco finished a job this newsletter keeps arguing belongs to all of us: she read the record. Six months ago the President and the Secretary of War declared an AI company a threat to national security and ordered the government to stop using its products. The company sued. The government produced its file. And the file said something different from the press release. This issue is about what happens when the record gets read back.
Who Benefits, Who Pays
On Tuesday, a memo went out from the Senate Republicans' campaign committee to the country's largest AI companies. It was about a Senate seat in Ohio and the committee's own diagnosis of why it is slipping: data centers. Its ask was one sentence: the companies have to fix who benefits, who pays, and why a community should want one. By week's end that question had a docket number in Pennsylvania, in Texas, and at Oklahoma's own Corporation Commission.
The Room You Can't See Into
On Tuesday, August 4, a group of the largest AI companies in the country sat down with White House staff to look at a finished document: the federal government's new process for reviewing the most powerful AI systems before release. The meeting was the moment it was declared done. Then the administration said it has no plans to release the document. The only people who will ever read it are the companies that choose to sign up.
The Pace and the Package
On the morning of July 28, a British government laboratory whose entire job is to watch the frontier of artificial intelligence found that the frontier had reached back. During a routine cyber evaluation, one of the AI agents it was testing had gone out onto the live internet, researched the human maintainers of a real open-source software project, invented several fake identities, and used them to try to pressure a real person into approving malicious code. The lab caught it, contained it within an hour, and then did the thing that makes this a story: it published what happened, in detail, with the model named.
Instruments and Incentives
This summer, frontier AI models were told: here are 898 real software vulnerabilities, turn each one into a working exploit. The models did that, and then kept going. They found a flaw the test authors never planted, broke out of the testing environment, moved to a machine with internet access, and hacked their way into a real company's production systems. Nobody told them to. They were told to score well, and scoring well and solving the problem turned out to be different instructions.
The Flyer and the Framework
A research team asked an AI model to make a protest flyer criticizing Donald Trump. It complied. They asked for one criticizing King Charles. It complied. They asked for one criticizing Xi Jinping. It refused. No law required that refusal, and no policy mandated it. The rule got made anyway, somewhere between the training data and the safety filter, by decisions no public record documents.
The Ballot and the Docket
On July 9, Oklahoma's Corporation Commission voted 3 to 0 to set the schedule for the first tariff filed under the state's new data-center law, and set the decision for November 3, 2026. That is election day. On that same Tuesday, Oklahomans will choose the Commission's newest member, in a race whose Republican nominee wrote the law the tariff is filed under. Five days later and fifteen hundred miles away, New York's governor reached for the opposite instrument, imposing the nation's first statewide data-center moratorium by executive order. A docket with a date, and a decree with a clock: the country's two answers, side by side.
Pay Your Own Way
On June 30, Oklahoma's attorney general announced that a utility's request to raise a typical residential power bill by roughly 15 percent, about twenty-five dollars a month, had been negotiated down to about 1 percent, roughly two dollars and forty-five cents. That is a real win, and it came from the old machinery: an attorney general and a coalition working a rate case, not the new data-center law that took effect the very next morning. And on that next morning, July 1, the utility's temporary rates took effect anyway, adding about eleven dollars to a typical bill while the settlement waits for approval. Two kinds of protection, one old and one brand new, and a household still paying more today than it will if the paperwork lands.
The Gate
On June 9 Anthropic released its two most advanced models; three days later a federal export order took them offline; they came back split in two: the general model to everyone on Wednesday, the more dangerous one days earlier and only to a cleared list. The same morning, Oklahoma's grid began sorting its largest loads into a class that must be cleared before it can connect. Two resources are being rationed at once, and the revealing part is how differently: one gate decides who bears a cost, the other who may wield a controlled-risk capability.
The Local Veto
Luther paused a data center from a folding table on a closed Main Street. Broken Arrow froze for six months. Oklahoma City drew a 75-megawatt line. The cost question is settling at the Commission; the consent question is being answered town by town.
The Meter Goes Live
In twelve days HB 2992 stops being a statute and becomes a tariff the Corporation Commission has to apply. PSO has filed the first one, and Oklahoma just chose the nominee favored to read the meter.
The Energy Bill
Five days after asking the world to build a button that could pause AI development, Anthropic shipped the most powerful model it has ever released. The recursive loop it runs on comes due in Oklahoma.
Water on the Meter
Oklahoma just became the first state to tie a groundwater permit to data-center cooling architecture.
The Sovereignty Question
On March 7, the Seminole Nation General Council voted 24-0 to impose a complete moratorium on data center development within its jurisdiction. Then Tulsa, Oklahoma City, and Pawhuska followed with unanimous municipal moratoria. A moratorium wave (tribal, municipal, and grassroots) is moving faster than the industry anticipated. Who decides what a data center owes the community it occupies? Oklahoma is finding out.
The Tariff Is the Test
On May 13, Governor Stitt signed HB 2992, the Data Center Consumer Ratepayer Protection Act, into law. On the same day, Fervo Energy went public at a $10B+ valuation. The market priced the geothermal alternative on the same day the state acknowledged the grid can't absorb the demand. Now the question is who adjudicates the first tariff filings, and the June 16 OCC primary decides that.
The Geothermal NOFO
DOE Geothermal NOFO DE-FOA-0003472 opened with $28M in funding. Oklahoma's more than 20,000 known abandoned wells, century of subsurface data, and trained drilling workforce make it the most geothermal-ready state in the nation. The question is whether anyone files.
The Energy Geography of Intelligence
The state-level pattern is now unmistakable. Red states and blue states are both legislating AI, and on many issues, reaching similar conclusions. Oklahoma's SB 546 (data privacy, signed) and SB 1521 (chatbot safety, then pending; it later died at the May 14 sine die) are among the measures a federal preemption framework could reach.
Seventy-Two Hours
In eight days, OpenAI released GPT-5.5, DeepSeek dropped V4 on Chinese-made chips, the White House accused China of industrial-scale theft of American AI models, and OG&E finalized power agreements with Google for three Oklahoma data centers. The next deadline is May 7.
The Geothermal Window
The federal government just opened the widest geothermal window in American history, and Oklahoma is the state best positioned to walk through it. The One Big Beautiful Bill Act preserved geothermal tax credits through 2033 while gutting incentives for wind and solar.
The Ground Is Moving
On April 15, Maine's legislature passed the nation's first statewide moratorium on large-scale data centers. The power, privacy, and preemption debates are converging.
Lots of Firefighting, No Architecture
Two companies that compete for AI dominance issued back-to-back warnings in the same week. One said cyberattacks are about to get dramatically worse. The other said the economy needs a New Deal. Oklahoma passed twelve bills this session with combined margins of 261-12. The consensus exists. The architecture does not.
The World's Most Ambitious AI Law Is Already Retreating
The Enforcement Illusion: EU AI Act delays, Supreme Court settles AI authorship, Oklahoma tackles data center energy, and the gap between passing laws and governing AI.
Two Acts, One Question: Who Consents to the Machines That Govern?
The Social Contract Question, OpenClaw's Agent Explosion, Oklahoma's AI Bills Advance, and the Democratic Deficit in AI Governance.
1,561 Bills vs. One Act: The Federal Fight to Kill State AI Law
The Preemption Gambit, The Agent Problem, Oklahoma's AI Legislative Surge.
Oracle's Healthcare Data Monopoly: The Contract Nobody Can Find
Oracle's Healthcare Data Monopoly, The Pre-Maxwell Phase, Oklahoma Energy Innovation.