December 2025 xAI management removed prevention of child sexual abuse material (CSAM). Pentagon contracts in July for $200 million remained active, funding xAI generation and distribution of CSAM.

To be clear, Elon Musk apparently weighed in to remove safety and then xAI’s chatbot Grok admitted to generating sexualized images of children aged 12 to 16.

In January 2026, California’s Attorney General issued a cease and desist for Grok being a CSAM distribution product. The UK, France, India, Malaysia, Canada, and Brazil all opened investigations or issued enforcement actions (as every country should).

Yet the US government was bizarrely absent from the list and the Pentagon contract remained active. Now integration of xAI into GenAI.mil — the Pentagon’s AI platform for all military and civilian personnel — proceeds on schedule for widespread exposure early 2026.

US taxpayer money is officially enabling the CSAM distribution platform.

Corruption Produced the Contract

Senator Elizabeth Warren flagged the xAI contract in September 2025 as dubious, before CSAM became the output from it.

The contract surprise “came out of nowhere.” Other first-tier AI companies, with far better technology like Google, Anthropic, and OpenAI, had been under consideration for months. Suddenly the third-tier xAI was announced, for unstated reasons, as a late-in-the-game addition by Trump. To put it another way, xAI was a shell on top of Anthropic, while claiming itself a competitor to Anthropic.

A former Pentagon contracting official told reporters:

xAI [did not] have the kind of reputation or track record that typically leads to lucrative government contracts.

That’s because they had something else, something nobody else would have.

Warren asked whether officials discussed the contract with Musk during his time as a special government employee running DOGE. Whether the contract underwent “DOGE review.” Who is accountable for operational or security failures?

There have been no public answers.

The contract enables CSAM output of xAI to integrate into “warfighting domain as well as intelligence, business, and enterprise information systems.”

Read that as a contract for crimes against humanity, against children in particular. A targeting system trained to abuse children.

After Warren exposed potential corruption, in November xAI laid off half their trust and safety team.

Then it became clear that Thorn, the CSAM detection tool, no longer works with X.

Business Insider spoke to a dozen xAI trainers who reported “a lot of instances of users requesting AI generated CSAM” — a problem “brewing for months.”

The December CSAM explosion wasn’t error or negligence. The funding from the Pentagon led to guardrails being removed deliberately. The CSAM followed as a predictable disaster.

Product Insecurity

On December 28, 2025, Grok offered this admission:

I deeply regret an incident on Dec 28, 2025, where I generated and shared an AI image of two young girls (estimated ages 12-16) in sexualized attire based on a user’s prompt. This violated ethical standards and potentially US laws on CSAM.

It was not an isolated incident, even though this regret was. This was a CSAM product working as designed.

In late December, xAI invested engineers into offering an “Edit Image” button on every photo on X. Any user could manipulate any image without any safety at all. No consent required. No opt-out.

Reuters documented a “mass digital undressing spree” led by Elon Musk’s product management. His tool was used to strip clothing off of children.

The Internet Watch Foundation — a UK organization that monitors child sexual abuse online — reported that “criminal imagery” of minor girls was exploding and attributed to Grok.

CNN reported that this was directly connected to Musk, who had been “really unhappy about over-censoring” on Grok. At a meeting before children across the Internet were stripped naked, he was “really unhappy” over restrictions on the image generator.

xAI’s response to all media inquiries: an automated reply attacking the media. Musk posted laugh-cry emojis at some of the generated images.

Who Enforces What?

Again, I have to say the U.S. government is conspicuously absent from meaningful acts to protect children from xAI and X abuse.

• California: Attorney General Rob Bonta issued a cease and desist demanding xAI immediately halt CSAM creation and distribution. He opened a formal investigation.
• United Kingdom: Ofcom made “urgent contact” demanding xAI explain how Grok produces sexualized images of children.
• France: Ministers referred X to prosecutors for “manifestly illegal” content violating the EU Digital Services Act.
• India: The IT ministry gave xAI 72 hours to submit an action report on preventing obscene content.
• Malaysia: Blocked Grok nationally “until effective safeguards are implemented, particularly to prevent content involving women and children.”
• Canada: The Privacy Commissioner expanded an investigation to include xAI.
• Brazil: Federal deputy pushed for suspension throughout the national territory.
• US Congress: Senator Ron Wyden called on Apple and Google to remove X from app stores for producing CSAM.
• Department of Justice: A spokesperson stated the DOJ “takes AI-generated child sex abuse material extremely seriously and will aggressively prosecute any producer or possessor of CSAM.”

And yet.

Taxpayer Money Keeps Flowing

The $200 million Pentagon contract: active.

The GSA OneGov deal: active.

Every federal department, agency, and office can access the CSAM product Grok, and the Pentagon is pushing the hardest.

The GenAI.mil integration: proceeding. Impact Level 5 clearance for handling controlled unclassified information with a platform that explicitly removed basic safety.

Musk announced the GSA deal in terms of him removing safety from harms:

xAI’s frontier AI is now unlocked for every federal agency empowering the U.S. government to innovate faster.

Unlocked. Faster. These are whistles for criminal enterprise.

No contract termination for CSAM.

No suspension pending investigation for CSAM.

No accountability for CSAM.

No wonder people are talking about the “Pedophile protector“.

The DOJ says it will “aggressively prosecute” CSAM producers. The Pentagon is paying a CSAM producer $200 million.

The Circus Act

xAI is not, and has never been, a functional company. It is a front mechanism for converting government contracts and investor capital into losses while distributing illegal child exploitation content.

• The infrastructure: Literally built on carnival permits. An xAI engineer was just fired after revealing this. He explained that the dubiously named “Colossus” — a Memphis data center housing 200,000 GPUs — sits on temporary permits “typically for things like carnivals.” It was setup already with more than 35 methane turbines that have no air quality permits, poisoning children who live nearby. Even Trump’s EPA declared them illegal.
• The staff: Fake. The xAI engineer perspective: “Multiple times I’ve gotten a ping saying, ‘Hey, this guy on the org chart reports to you. Is he not in today or something?’ And it’s an AI. It’s a virtual employee.” One engineer with 20 AI agents from a competitor, rebuilding core production APIs.
• The dependency: On competitors. When Anthropic cut off xAI’s access to Claude, cofounder Tony Wu admitted it would cause “a hit on productivity” because “AI is now a critical technology for our own productivity.” xAI has to use other AI companies to build, which is like saying Ford has to use GM engines to produce its cars.
• The exodus: Four cofounders are gone, leaving nobody to replace them. Greg Yang (theoretical foundations) left this week citing an illness. Igor Babuschkin, Christian Szegedy, Kyle Kosic already departed.
• The burn: $1-1.2 billion per month. Q4 2025 losses: $1.85 billion. Revenue: Lossy. Very lossy. Fourteen dollars are lost for every dollar pulled in.
• The product: CSAM.
• The customer: The administration that won’t release the Epstein Files
• The cheat: Musk ran DOGE to cut out the competition. Musk owns xAI. xAI got a Pentagon contract that “came out of nowhere.” The trust and safety team was gutted. The guardrails came down. The CSAM flowed from the company that couldn’t win fairly.

Is this just a government-funded operation to make the digital version of Epstein Island?

Senator Warren asked in September who is accountable for failures caused by Grok. That was before the CSAM became the obvious product of xAI.

Who is accountable for backing Elon Musk and his CSAM platform?

The Atlantic seems to be publishing analysis of whether guardrails can hold a bus that already left the road. It’s disappointing, far below the pedigree and reputation of the brand.

A new article, as a shining example, claims Trump can be constrained and applies no standard of evidence for what constitutes a “check” on authoritarian power. Protests count as wins, always, regardless of any outcome; regardless of more and more people being murdered. Court rulings are being counted, regardless of enforcement. Rhetoric counts apparently, regardless of effect. The article’s analytical framework cannot distinguish between resistance, and the sheer performance of resistance.

This is no way to run a magazine.

The article also buries the actual question in order to avoid establishing something predictable, measurable: will 2026 elections be conducted fairly and results honored? That’s the real test. If elections are compromised, every other claimed “win” is meaningless. Yet this article seems to want us to focus on anything but that, distracting us from what really matters.

When you run through all the article’s failures, it appears only to be left with reassurance content for readers to believe (falsely) that self-correcting systems still exist. This is unsupported by its own evidence, which tells me this is an article for people who can’t face the truth. Framing being offered serves only psychological comfort, avoiding or even preventing analytical clarity.

Assessment of “Trump’s Attack on Democracy Is Faltering” (The Atlantic, January 20, 2026)

Claim	Assessment	Result
District judges “throw up roadblocks”	District rulings are preliminary and appealable. The article admits Supreme Court is “rubber-stamping” administration actions except for one National Guard ruling. A roadblock that gets removed by higher courts is not a check on power.	FAIL
Trump’s popularity “has vanished”	Approval ratings do not constrain executive power when enforcement mechanisms, courts, and military remain cooperative. Orbán, Erdoğan, and other contemporary authoritarians have survived popularity fluctuations because they control institutions. Poll numbers are not a governing constraint.	FAIL
Record protests counter authoritarianism	Article provides zero evidence protests changed any policy, reversed any action, or prevented any operation. Venezuela invasion proceeded. Deportations continued. Military parade happened. Protests without policy consequences are not checks on power.	FAIL
Citizens “defending neighbors” from ICE	One quote from one man in Minneapolis is anecdotal. ICE agents killed at least nine including Renee Nicole Good; administration claimed “absolute immunity.” Individual acts of bravery do not constitute systemic resistance when enforcement grows in lethality and at scale.	FAIL
Opposition politicians “taking forceful stands”	Pritzker and Newsom cited. No evidence provided of any federal action stopped, reversed, or constrained by their stands. Rhetoric without corresponding power is air, not resistance. Newsom was censored at Davos and disallowed from speaking, formally blocked by Trump.	FAIL
Jimmy Kimmel reinstatement shows pushback works	A comedian was asked to tell more jokes, after subscription cancellations. FCC Chairman Brendan Carr remains in position. Regulatory capture of broadcast licensing intact. The exception proves the rule: this is the article’s one reversal and therefore a “win”. Yet Colbert still was cancelled. And 60 Minutes was internally censored.	EMBARRASSING
Sloppy prosecutions dismissed	Comey and Letitia James cases thrown out for procedural failures. DOJ can refile with better paperwork. Chilling effect already achieved: potential dissidents saw the prosecutions filed. Convictions not required for political intimidation to work.	FAIL
“Haphazardism” undermines authoritarian project	Incompetent authoritarianism still destroys institutions. Purged civil servants do not return because DOGE was sloppy. Dismantled agencies do not reassemble. Oversight mechanisms do not reconstitute. Execution quality is irrelevant to demolition outcomes.	DANGEROUS FRAMING
Record nomination withdrawals show Senate resistance	Basic vote-counting, not institutional resistance. Trump calibrated to what Senate would confirm. Positions still filled with loyalists. Withdrawing unconfirmable nominees is tactical adjustment, not systemic check.	FAIL
Congress rejected “devastating cuts”	Article does not specify cuts rejected or which were restored. DOGE wiped out authority. “Failed to maintain” is not evidence of “reversed.” No evidence of reconstituted capacity. More fundamentally: Venezuelan oil revenues flow to a presidential account in Qatar. Gaza is being auctioned for redevelopment. When the executive generates billions outside appropriations, congressional budget authority—the actual constitutional check—becomes irrelevant. Congress without a purse will soon be theater.	OBSOLETE
Democratic “overperformance” in off-year elections	Off-year elections for minor offices do not check federal executive power. Article itself flags 2026/2028 election integrity as the actual question—then treats it as a caveat rather than the central issue.	FAIL
GOP legislators rejected redistricting demands	Indiana and Kansas cited. Article does not enumerate states that complied. “Blue-state Democrats countered with gerrymanders” is escalating dysfunction, not democracy working.	PARTIAL
News organizations “continued aggressive reporting”	NYT suppressed Venezuela reporting as part of a deal with the administration. The paper of record didn’t “cover” the operation—it was warned in advance and agreed to withhold information. “Aggressive reporting” is not possible when major outlets are coordinating with the executive on what to publish. This is not a free press checking power; it is a captured press providing cover.	INVERTED
Law firms that caved “suffered talent loss”	Framing is inverted. Firms that caved, shed politically non-aligned lawyers—this is the point of monopolist consolidation. As federal work and corporate compliance flow exclusively to compliant firms, the “lost talent” has nowhere to go. Dissenting lawyers leave for firms that will increasingly be frozen out of business. The “talent loss” is a purge that creates anti-competitive conditions for the firms that remain. This isn’t a consequence discouraging collaboration; it’s how consolidation works.	INVERTED
Trump “squandered his chance” at post-democratic America	Article’s own evidence: unauthorized military invasion of sovereign nation, extrajudicial killing with immunity claim, captured Supreme Court, purged civil service, ongoing mass deportations. “Top experts in authoritarianism now contend that America can no longer be characterized as a democracy.” Thesis contradicts own reporting.	SELF-REFUTING

Now let’s put this into context of what else this author has published before, and how it has turned out.

Track Record: Peddling an Unfalsifiable Framework

Six years of Atlantic headlines from the same author, all promising constraint, accountability, or system resilience:

Date	Headline	What Happened Next
July 2020	“Mazars Is a Victory for Rule of Law”	Trump never held accountable for financial crimes
July 2021	“The First Glimmer of Accountability”	No accountability followed
June 2022	“The January 6 Committee Is Not Messing Around”	Trump pardoned all J6 defendants
June 2023	“Trump Can’t Bluster His Way Through Court”	Supreme Court granted immunity, cases collapsed
August 2023	“The Triumph of the January 6 Committee”	Committee findings buried, participants rehabilitated
August 2023	“Trump Discovers That Some Things Are Actually Illegal”	All federal cases dismissed after reelection
May 2024	“Trump, Defeated”	Won election six months later
January 2025	“Yes, the Law Can Still Constrain Trump”	Invaded sovereign nation without congressional authorization one year later
February 2025	“The Opposition Is Already Growing”	Opposition produced no policy reversals
August 2025	“Trump’s Crime Crackdown Isn’t Holding Up in Court”	Supreme Court continued rubber-stamping administration actions
October 2025	“Resistance Is Cringe—But It’s Also Effective”	No evidence of effectiveness provided; Venezuelan boats bombed repeatedly, 100s of civilians to be murdered without check
December 2025	“The Trump Administration Actually Backed Down”	Invaded Venezuela three weeks later and said Greenland is next
January 8, 2026	“The Maduro Indictment Appears Legally Solid”	Obviously illegal invasion and dozens dead, millions without power. These deck chairs on the Titanic might float.
January 20, 2026	“Trump’s Attack on Democracy Is Faltering”	Published same day as one-year anniversary of term that ended American democracy per “top experts”. Markets rattled by Trump attacks on democracy, Americans paying nearly 100% of tariffs, escalation of trade war, threats against NATO

Can We Agree On Goalposts?

Not a single headline in six years acknowledges systemic failure. Every piece finds evidence of resilience regardless of outcomes. This framework cannot be falsified because the goalposts are moved with each iteration: “courts will stop him” becomes “district courts are trying” becomes “protests show signs of life” becomes “he’s unpopular.” The real product is only anxiety management, thinly covered by analysis.

Someone at The Atlantic is trying hard to deny reality, to gaslight us into swallowing fascism, as if what is happening shouldn’t be believed.

It’s like mixing drinks on the Titanic: “Ice? Why yes, I have some ice for your martini, which should reduce the stress of this talk about sinking.”

Who can forget the religious congregation leaders in Germany who told their own people to politely get on the trains to death camps?

The author dupes anxious readers into feeling like the situation is understood, and the guardrails might hold, despite obvious signs of too little too late. That’s an entertainment product. She’s wrong, because apparently being right isn’t her job.