joe curlee

We are watching the death of personal technology, the First Amendment, and the internet in one fell swoop. Whether this is planned by a person by Elon Musk, who has in the past signaled that he would like to fully control online speech through both his actions (purchasing Twitter and reshaping the algorithm) and his words ("That's done" in response to people posting anonymous inflammatory statements online, and "freedom of speech, not freedom of reach"—the cry of any true statist). The federal government strongly backs AI companies. ChatGPT is in bed with the NSA, xAI is in bed with the Department of War, etc.—each is a data-gathering factory for the federal government. Your conversations are monitored and flagged for anything suspicious on many of these platforms, including Anthropic's Claude. Anything suspicious is forwarded directly to law enforcement agencies. The moves these AI companies make, legally grey or outright illegal, are shielded by the federal government in the name of "national security" and "defeating China." In reality, any time a person or entity is forcing you to move fast (think 9/11 reactions, which rushed us into a 20-year war), the person or entity doing the rushing is manipulating you down the path they want you on. Both sides of the government are working in collaboration to move us towards a path of artificial technological scarcity, which opens the door to censorship, as you will see in the article below, and to remove our First Amendment rights from the internet. This current frenzy has all the hallmarks of a false flag and psychological operation. The following article is AI-generated. I directed the research using Codex GPT-6.1 Sol. I used this model to create an outline from the research. I then wrote this intro that you're reading and asked it to write the research article for me. I have reviewed it prior to posting this blog post.

Start with the hardware. In his late-October 2026 interview on Cheeky Pint, Sam Altman discussed the desire to run open models locally. He argued that allocating scarce DRAM to personal computers with low utilization was “quite terrible” and advocated “maximum utilization in the cloud.” His stated rationale was efficiency during a supply shortage. Original interview transcript, around minute 29.

Utilization is one consideration. Ownership, privacy, resilience, and independence are others. Memory sitting in my computer gives me the ability to run software without sending every interaction to a provider. Its value cannot be measured entirely by how continuously a corporate data center could use it.

Altman's preference matters because his company has the scale to influence suppliers. OpenAI's October 2025 announcement with Samsung and SK described plans to expand advanced-memory production toward 900,000 DRAM wafer starts per month. That announced target illustrates the scale at which these companies negotiate access. OpenAI's memory partnership announcement.

Micron made the priorities even more explicit. In December 2025, it announced its exit from the Crucial consumer business, tying the decision to AI-driven demand and improving supply for larger strategic customers. Consumer-channel shipments were scheduled to end in February 2026. Consumers did not imagine this shift. A major manufacturer announced it. Micron's announcement.

TrendForce's September 2026 research describes suppliers prioritizing advanced production capacity for server products while graphics-memory supply remains constrained. It forecast further increases in conventional DRAM contract prices during the fourth quarter. The pressure on other buyers follows a business decision about whose demand is most profitable to serve. TrendForce's research.

Some scarcity is physical: factories, packaging capacity, electricity, and production time impose real limits. The political and commercial question is who gets access within those limits—and whether contracts, software dependencies, and regulation make independent alternatives harder to sustain. That is what I mean by artificial technology scarcity: restrictions on practical access created or intensified by institutional choices.

The price of owning powerful consumer hardware is part of that story. Nvidia launched the RTX 4090 at $1,599 in 2022 and the RTX 5090 at $1,999 in 2025, an approximately 25 percent increase in the nominal flagship starting price. These are different generations of hardware, but the cost of buying into Nvidia's top consumer tier rose by $400 before any retailer markup. RTX 40-series launch prices, RTX 50-series launch prices.

Nvidia's position extends beyond selling chips. France's competition authority identified expensive computing inputs and the company's proprietary CUDA ecosystem as barriers in generative AI. Software built around one vendor's hardware makes changing suppliers a substantial engineering decision. A competitor offering a chip does not automatically give customers an inexpensive way out of the surrounding ecosystem. The authority's English report, particularly paragraphs 124–130.

The cloud partnerships add another layer. The FTC's January 2025 report examined Microsoft–OpenAI, Amazon–Anthropic, and Alphabet–Anthropic arrangements. It documented cloud-spending commitments linked to investments, varying control and exclusivity rights, discounted computing access, and potential switching costs. These structures can turn an investment into a long-term infrastructure dependency. The FTC identified competition risks arising from the arrangements. FTC findings.

These concerns have a legal framework. Section 1 of the Sherman Act addresses agreements that unreasonably restrain trade; Section 2 addresses monopolization and attempted monopolization. Market dominance alone does not establish an unlawful monopoly: the acquisition or maintenance of monopoly power through anticompetitive conduct matters. Section 7 of the Clayton Act separately addresses acquisitions whose effect may substantially lessen competition or tend to create a monopoly. Higher GPU prices, large investments, and market share are reasons to investigate the structure and conduct of these businesses. They do not, by themselves, prove a violation. Investigators should examine the contracts, exclusivity, switching barriers, and effects on competition. DOJ's explanation of the antitrust laws, Clayton Act Section 7, 15 U.S.C. § 18.

Musk has publicly predicted where financing pressures could lead. His October 25 remarks forecast “guppy-level death and consolidation” among specialized AI cloud providers, contrasting their position with SpaceX's revenue and access to financing. His own corporate consolidation is already concrete: SpaceX announced its acquisition of xAI in February 2026. Independent providers face financing pressure while the largest businesses can draw on whole corporate empires. Reported remarks, the acquisition announcement.

The government's role deserves the same attention. Biden promoted rapid AI infrastructure development in January 2025. Trump subsequently pursued his own infrastructure agenda. Their policies differ, but enthusiasm for large AI buildouts crosses administrations and party lines. Biden's January 2025 statement.

Trump's July 2025 data-center order directed an initiative for financial support that could include loans, guarantees, grants, tax incentives, and purchase commitments. Qualification routes included projects committing at least $500 million, projects adding more than 100 megawatts, and national-security or discretionary designations. Washington was organizing federal support around this buildout. The scale of the qualifying projects tells us something about the kind of development it wanted to accelerate. The executive order.

In December 2025, another order directed the creation of a federal task force to challenge state AI laws inconsistent with the administration's policy and sought funding leverage against certain state restrictions. The order's purpose includes a minimally burdensome national framework. The industry's freedom from regulatory burdens had become an explicit objective of federal intervention. Executive Order 14365.

An executive order directing legal challenges and federal agency action does not, by that direction alone, repeal every state AI law. The order also calls for a legislative recommendation for a federal framework. Whether a particular state requirement is preempted requires an actual legal basis, not simply an administration's preference for fewer restrictions. Executive Order 14365, particularly Sections 3 and 8.

Programs for wider access also exist. NSF's NAIRR initiative aims to expand research computing access and describes pathways for startups and small businesses. The question is whether smaller organizations can obtain enough durable, independent capacity to compete—not whether every public program ignores them. Assistance that leaves an organization permanently dependent on a dominant provider should be evaluated with that dependency in mind. NSF's NAIRR program.

The military and intelligence relationships are public. OpenAI appointed former NSA director Paul Nakasone to its board in June 2024. xAI announced its selection for the military's GenAI.mil platform in December 2025, including access to insights from X. The institutions building consumer AI are also building relationships at the center of American military and intelligence power. OpenAI's appointment announcement, xAI's military partnership announcement.

OpenAI's own account of its Pentagon agreement says something more specific: its March 2026 update describes prohibitions on domestic surveillance of U.S. persons, including through commercially acquired personal information, and says intelligence agencies such as the NSA would require a new agreement to use its services. Those are the company's published representations. They belong in any honest assessment, alongside demands for enforceable safeguards and independent oversight. OpenAI's agreement and update.

Anthropic's dispute with the Pentagon shows the pressure those relationships can create. In February 2026, Dario Amodei said the company had already deployed Claude across military and intelligence uses but would not accept removing restrictions concerning mass domestic surveillance and fully autonomous weapons. His statement described threats involving a supply-chain-risk designation and the Defense Production Act. Here was an AI company publicly describing government pressure over the limits it wanted to maintain. Its resistance matters. So does the leverage it said the government threatened to use. Amodei's statement.

The Defense Production Act makes the allocation question concrete. Under 50 U.S.C. § 4511, the President has authority to prioritize national-defense contracts and allocate materials, services, and facilities, subject to statutory conditions. Section 4511(b) specifically limits controlling a material's general civilian-market distribution without the required findings. But national-security language is not blanket antitrust immunity. Section 4558(a) expressly rejects a general antitrust shield under the Act, while subsection (j) provides a conditional defense for qualifying authorized agreements and actions. Anyone claiming protection should identify the actual authority and its limits. 50 U.S.C. § 4511, 50 U.S.C. § 4558.

The underlying concern survives those distinctions. A government that becomes a major customer of concentrated technology providers gains relationships and potential leverage. A public pushed into the same providers' systems has fewer practical ways to keep its activity outside those relationships. That is a structural risk worth confronting without inventing a secret data-sharing arrangement.

Consumer chat monitoring is directly documented. OpenAI says automated systems analyze user content and behavior, with flagged activity assessed by trained personnel. Its April 2026 explanation says it notifies law enforcement when conversations indicate an imminent and credible risk of harm to others. The company therefore operates a system capable of moving a conversation from automated analysis to human investigation and, in qualifying cases, outside disclosure. OpenAI's monitoring and referral explanation.

Anthropic's consumer retention policy says material flagged as violating its usage policy can be retained for up to two years, with trust-and-safety classification scores retained for up to seven. Its government-request policy describes disclosure under valid legal process and an exception for emergencies involving imminent physical harm or death. A conversation can leave behind a safety classification that outlasts the retained conversation itself. Users should know that before treating a chat window as a private place to think. Consumer retention policy, government-request policy.

Federal disclosure law also requires distinctions. For covered electronic-communication and remote-computing services, the Stored Communications Act generally restricts disclosure and provides specific exceptions. One permits disclosure when a provider believes in good faith that an emergency involving danger of death or serious physical injury requires it without delay. Other exceptions include lawful consent; a separate section addresses compelled disclosure under legal process. The applicable protection and process depend on the service and information involved. These provisions do not establish a general duty to forward every chat labeled suspicious directly to police. 18 U.S.C. § 2702, including subsections (b)(8) and (c)(4), 18 U.S.C. § 2703.

There are separate mandatory reporting rules. Section 2258A requires covered providers with actual knowledge of specified apparent child-sexual-exploitation offenses to report to NCMEC's CyberTipline. Subsection (f), however, says that section does not require monitoring users or affirmatively searching, screening, or scanning their communications. A company's monitoring policy, an emergency referral, a mandatory CyberTipline report, and compliance with a warrant are different things. We should demand a clear account of which process is being used and why. 18 U.S.C. § 2258A(a) and (f).

I understand the argument for responding to credible threats. I also want people to understand the arrangement before pouring private thoughts into a conversational interface. A provider can hold both the conversation and a classification of it. Its systems and judgments become part of the boundary between private thought, account enforcement, and possible outside intervention.

Commercial data's intelligence value is already acknowledged. A declassified advisory report prepared for the Director of National Intelligence describes government acquisition of commercially available information and the privacy risks created by its sensitivity, aggregation, and possible reidentification. The intelligence community understands the value of assembling commercial information about people. We should understand the civil-liberties stakes of that capability just as clearly. The declassified report.

Age-verification laws create a separate route to dependence and exclusion. Bluesky initially blocked Mississippi in August 2025, citing the resources required to comply. Its December update reopened access for adults completing age assurance while continuing to exclude minors. The company explicitly said the compliance demands disproportionately harmed smaller platforms. Regulation had affected whether people could use a competing service at all. Bluesky's explanation and updates.

Florida provides another example. Section 501.1736 applies to platforms meeting specified usage, algorithm, and feature criteria; it prohibits accounts for children under 14 and requires parental consent for ages 14 and 15. It is not a rule covering every website or a command that every adult upload government ID. On October 14, 2026, the Eleventh Circuit reversed the preliminary injunction as to the challengers' as-applied First Amendment claim, vacated it as to the facial challenge, and remanded. The panel treated the restrictions as content-neutral and concluded that the as-applied challenge was unlikely to succeed under intermediate scrutiny. This was a preliminary-injunction decision, not a final judgment resolving every constitutional issue. Florida Statutes § 501.1736, the court's opinion, especially pages 23–33 and 47.

The Supreme Court has also upheld an age-verification requirement. In Free Speech Coalition v. Paxton, decided June 27, 2025, the Court held that Texas's requirement for covered commercial sites distributing sexual material obscene to minors survived intermediate scrutiny because it only incidentally burdened adults' protected speech. That ruling must be confronted directly. It concerns a particular restriction on minors' access to sexual material; it does not decide the constitutionality of every identity check for political discussion or general social-media use. Courts examine the law, the speech, and the burden involved. The Supreme Court's opinion.

Age verification does not invariably require uploading government ID. Methods differ, and privacy-preserving design matters. Ofcom, for example, lists methods including photo-ID matching and facial age estimation. But any required access check deserves questions about who handles the information, what is retained, how mistakes are challenged, and whether smaller services can afford compliance. Child protection is a serious objective; attaching that label to a requirement does not settle whether the requirement is proportionate. Ofcom's explanation.

The same struggle over control is familiar in entertainment. Nvidia's GeForce NOW FAQ gives Performance and Ultimate subscribers 100 monthly hours, with limited rollover and options for additional playtime. Sony sells PS5 Pro's disc drive separately and requires an internet connection to pair it during setup. The drive supports DVDs and Blu-rays, but even adding physical-media capability involves contacting the provider. These business decisions normalize paying for remote access and authenticating capabilities on hardware we buy. Nvidia's FAQ, Sony's pairing requirement, supported media.

Each added dependency gives an institution another decision to make about access. Whether the subject is a model, a game, or public conversation, ownership and the ability to operate independently deserve weight alongside convenience.

The history of broadcasting helps explain the stakes, provided we get it right. Spectrum scarcity was an important rationale in Red Lion for broadcast-specific obligations. Pacifica also relied on broadcasting's pervasive presence and accessibility to children. Scarcity was never the only justification. In Reno v. ACLU, the Supreme Court distinguished the internet from broadcasting and struck down the challenged online-indecency provisions. Red Lion, Pacifica, Reno.

Even broadcasting does not give the FCC unlimited censorship power: 47 U.S.C. § 326 expressly withholds a general power of censorship under the Communications Act. And making GPUs or cloud access scarce does not itself repeal the First Amendment or transform the internet into licensed broadcast spectrum. The risk I am describing is control over practical access to computing and speech. The constitutional limits on government still have to be defended and enforced. 47 U.S.C. § 326.

The First Amendment constrains government; it does not guarantee that a private platform will distribute every post. In Moody v. NetChoice, the Supreme Court explained that selecting and presenting third-party posts can involve protected editorial judgment. It remanded the facial challenges for the required analysis rather than declaring every platform regulation invalid. Government coercion is a different matter: in NRA v. Vullo, the Court reaffirmed that officials cannot coerce private intermediaries to punish or suppress disfavored advocacy, allowing the NRA's plausibly alleged claim to proceed. Concentrated private control and unconstitutional government pressure are distinct problems, and both deserve attention. Moody v. NetChoice, NRA v. Vullo.

Anonymous political speech has constitutional protection of its own. In McIntyre v. Ohio Elections Commission, the Supreme Court struck down a law requiring identifying information on campaign literature. The state could not justify that broad identification requirement with its asserted interests in preventing fraud and informing voters. That does not settle every online age check. It does establish that speaking anonymously about politics is a protected freedom, and that compulsory identification demands constitutional scrutiny. McIntyre v. Ohio Elections Commission, 514 U.S. 334 (1995).

That distinction makes independent infrastructure more valuable. A constitutional protection is harder to exercise when reaching people depends on a narrow set of providers, when leaving those providers is prohibitively expensive, or when participation requires satisfying an access regime that excludes you. The threat I am describing is the erosion of practical freedom while legal protections remain on paper.

We should demand competition scrutiny of hardware and cloud arrangements, interoperable software, and public computing investments whose benefits can be measured beyond the largest buyers. We should challenge disproportionate identity requirements, require meaningful limits on data retention and government access, and defend the ability to publish and run software independently.

We should also reject the demand to suspend scrutiny because the industry says the race is urgent. Leaders asking for exceptional support and fewer constraints owe the public more accountability, not less.

I want affordable local computing, strong encryption, anonymous participation, and continued offline access to purchased media. I want government restrictions tested against constitutional limits and corporate claims tested against their actual contracts and conduct.

I want a computer I control and an internet I can use without asking a corporation or the state for permission.

Updated November 1, 2026, to add the relevant antitrust, disclosure, and constitutional authorities and clarify the age-verification rulings.

• AI
• technology
• privacy
• social media
• hardware