China's bot farm found a new target: the power grid fight in your county.

Access, capital, and speech in American AI now run through deals nobody outside the room gets to see.

X pulled down roughly 200 Chinese accounts operating inside a 200,000-account network built to turn Americans against AI data centers and blame them for rising electricity bills, and OpenAI banned a separate batch of China-linked accounts running the same play. In the same stretch, the federal government demonstrated it can force a company to take models offline with 90 minutes' notice, punish firms that refuse to build weapons systems, and skip the rulemaking process entirely until a judge or a senator makes it stop.

What is taking shape here is not a policy debate about innovation versus safety. It is a market structure where enforcement works through back-channel pressure and procurement bans, where the rules that govern access are written in private commitments rather than public process, and where foreign adversaries have figured out that the softest target in American AI policy is a local land-use fight. The companies building these tools understand all of it. That is why an Anthropic-linked PAC just spent half a million dollars in a Senate primary.

01

China found the cheapest way to slow American AI — the county zoning hearing

X flagged roughly 200 Chinese accounts operating inside a network of about 200,000 fake profiles, all pushing Americans to oppose AI data centers and to blame them for rising electricity bills. OpenAI separately banned a batch of China-linked ChatGPT accounts running a similar operation aimed at U.S. audiences. Both landed in the middle of a domestic fight that was already ugly, where the arguments run from national security to grid strain to what a hyperscale facility actually does to a small community.

Why it matters: Foreign interference used to mean the presidential race. This is a county commission meeting. The operational problem is not that the bots are persuasive, it is that they say roughly what genuinely angry residents are already saying, which makes every real complaint easier to wave off as astroturf. Anyone organizing against a data center now has to establish they are actually local before they get heard, and anyone defending one has a ready-made reason to dismiss the opposition. That is the damage: not that Americans get fooled, but that legitimate local opposition gets cheaper to ignore.

Read the full story →  Fox News

 

02

Frontier AI is now regulated by phone call, not by rule

In June 2026 the federal government used its export control authority to force Anthropic to pull two models offline with 90 minutes' warning. The order was reversed 20 days later, after the company agreed to hand over pre-release access and commit to ongoing security reporting. By August 1 the administration had blown past its own 60-day deadline to publish the benchmarking framework meant to govern situations exactly like that one, leaving companies with no formal process to point to and bipartisan senators publicly demanding answers about both the legal authority and the standards.

Why it matters: The United States is not regulating frontier AI through public rulemaking right now. It is regulating through informal bilateral deals cut behind closed doors. That gives federal agencies the ability to pull a model offline with no advance notice, and it builds precedent out of commitments nobody outside the room can read. For the companies it is a mess of mixed signals: frameworks that are technically voluntary but carry the threat of criminal penalties. For customers there is no notice and no recourse. The exposure for a campaign is easy to state and hard to solve, because the tool your workflow depends on can disappear overnight and your contract will not have a word to say about it.

Read the full story →  Spencer Fane

See how it works →