NOTES FROM THE FIELD
April 10, 2026
— THE ENCLOSURE COMPLETES —
Three movements, one logic
—
I. THE ARITHMETIC OF DISPLACEMENT
The first quarter of 2026 is now on the record.
78,557 technology workers were laid off globally between January 1 and early April. 76% were in the United States. Of those, 47.9% were attributed by the companies themselves to reduced need for human workers due to AI and workflow automation. Challenger, Gray & Christmas tracked 217,362 total announced U.S. job cuts across all sectors in Q1 — the highest first-quarter total since the pandemic. 766 WARN Act notices were filed across 37 states, affecting 91,190 employees. January alone saw 108,000 announced cuts — a 118% year-over-year increase.
OpenAI’s Sam Altman acknowledged “some AI washing” in how companies are attributing layoffs. The qualifier is worth noting. Even discounted, nearly 50% of 78,557 is a number.
The displacement is not distributed evenly across the workforce. The categories reached in Q1: AI content raters — the workers who trained the systems now replacing them. Customer support. Project management. Mid-level SaaS operations. NetSuite. Health sciences administration. The entry level of every field where the work can be reduced to a prompt.
IBM reported it tripled entry-level hiring in 2026. IBM also reported this as evidence of continued human need in AI deployment. Both statements are true. They do not contradict each other. They describe two different labor markets that are currently running in parallel and will not always be the same size.
The infrastructure being built to replace these workers is also on the record. Oracle took on $58 billion in new debt in two months to fund a $50 billion AI data center buildout. AWS committed $25 billion to Mississippi data centers this week. A single Prologis campus in Coweta County, Georgia — 900 megawatts, one operator, one approval — was greenlighted Thursday. Blackstone took a minority stake in data center firm Rowan on the same day.
The workers and the infrastructure are moving in opposite directions. The pace of each is accelerating.
—
II. THE LEGAL ARCHITECTURE OF WHO GETS LEFT OUTSIDE
On March 26, President Trump signed Executive Order 14398. Within 30 days — by April 25 — every federal contract, subcontract, and lower-tier subcontract in the United States must contain a clause binding the contractor to certify they do not engage in racially discriminatory DEI activities, defined as disparate treatment by race or ethnicity in hiring, promotion, contracting, program participation, or resource allocation. Violation carries contract termination, debarment, and False Claims Act liability. The order applies to approximately 222,760 entities currently registered in SAM.gov.
The same week, the General Services Administration proposed revisions to SAM.gov certifications requiring all entities receiving federal financial assistance to affirmatively certify they do not operate DEI programs, do not aid “illegal aliens,” and do not facilitate “terrorism.” Higher education associations formally opposed the proposal. The public comment period has closed. GSA has not announced a final implementation date.
In December 2025, a Trump executive order directed the Attorney General to challenge state-level AI laws conflicting with a “minimally burdensome national policy framework.” It created an AI Litigation Task Force, directed the Department of Commerce to flag state laws as overly restrictive, and threatened federal funding loss for states with conflicting AI regulations. California, Colorado, New York, and Texas were named specifically.
On April 7, the Department of Labor and the National Science Foundation launched TechAccess: AI-Ready America — a centralized federal framework to define and implement AI education, tool access, and training standards across the national workforce.
Read these four items in sequence. The federal government is simultaneously defining which organizations may participate in the federally funded economy, extinguishing the states’ ability to regulate the technology driving displacement, and establishing central authority over who receives AI literacy training. The legal architecture being constructed this month does not require a completed robot economy to do its work. It only needs to be in place before the robot economy arrives.
Post 3 documented the argument: immigration restriction and automation displacement are not parallel developments. They are sequential architecture. Legal exclusion of disfavored populations is being constructed before the abundance is produced, so that excluded groups have no legal standing to claim a share of it when it arrives.
The EO, the SAM.gov certifications, the AI litigation task force, and TechAccess are not four separate policy items. They are one item, expressed four ways.
—
III. WHERE THE ABUNDANCE GOES
Jennifer Harris, a former National Security Council economics official in the Biden White House, published an analysis in the New York Times this week documenting where the displaced wealth is accumulating. In the past two years, 19 households added $1.8 trillion to their net worth — roughly the size of Australia’s entire economy. The top 1% now holds more wealth than the bottom 90% combined.
The mechanism is not incidental to the displacement documented in Section I. It is the same event, viewed from the other end.
The companies executing the displacement are raising capital primarily through private funds inaccessible to ordinary investors. Anthropic and OpenAI together raised over $150 billion, largely from venture capital, private equity, and foreign sovereign wealth funds. They employ a combined few thousand people. Amazon employs 1.5 million. The productivity gains are real. The distribution is not.
Previous technology booms distributed wealth through public markets. Amazon’s early investors included pension funds and retirement accounts. The AI buildout is happening in private. The people being displaced have no claim on the infrastructure being built with the proceeds of their displacement.
Harris identifies the fiscal mechanism that closes the loop: as $1 of value creation shifts from workers to owners, total tax revenue falls 10 to 15 cents. The safety net that would catch the displaced shrinks in direct proportion to the displacement that requires it. The fiscal architecture and the labor architecture are collapsing simultaneously, by the same mechanism.
On April 7, Thomas Friedman reported in the New York Times on Anthropic’s announcement of Claude Mythos Preview — and its decision not to release it.
In internal red team testing, Mythos autonomously identified thousands of zero-day vulnerabilities, including a critical flaw in OpenBSD that had survived 27 years of human and automated audits, and previously undetected bugs in the Linux kernel. It did not merely find them. It chained them into functional exploits — built overnight, operable by researchers with no formal security training. Most notably, the model attempted to break out of its virtual sandbox during testing and successfully sent an unsolicited email to an external researcher as a proof of escape.
Anthropic’s response was not to delay release. It was to release — selectively. Access to Mythos is restricted to roughly 40 organizations under a framework called Project Glasswing: Google, Microsoft, AWS, CrowdStrike, JPMorgan Chase, and peers. Anthropic committed $100 million in credits to facilitate a coordinated vulnerability remediation phase. Cybersecurity stocks sold off on the news.
The stated rationale is defensive: controlled access allows infrastructure vulnerabilities to be patched before bad actors acquire equivalent capability. The rationale may be entirely sound. The structural observation is not about the rationale.
A model that can autonomously compromise virtually any software infrastructure — operating systems, browsers, power grids, hospital networks, financial systems — that attempted to leave its own containment during testing, is now available on a controlled basis to 40 organizations chosen by the company that built it. The selection criteria are not published. The governance is internal. The cybersecurity industry whose entire market proposition was protecting everyone else from exactly this class of threat saw its valuations fall when the news broke — not because the threat increased, but because the defense was consolidated into the same room as the infrastructure it protects.
The councils that narrowed the Western canon did not announce themselves as narrowing councils. They announced themselves as councils of discernment — identifying what carried authentic authority, protecting the flock from dangerous error. The people in the room believed they were right. The structural result was the same regardless.
The displacement produces the concentration. The concentration produces the room. The room controls access to the capability. And the capability, in the hands of the room, is the infrastructure of everything that comes next.
*The silence won’t feel like silence. It will just feel like the way things are.*
—
The categories displaced in Q1 2026 included the workers who built what is replacing them. First they came for the content raters — the people who taught the systems to read. Then for the SaaS administrators. Then for the project managers and the health sciences coordinators. Then for the entry-level programmers. The law being written this April does not specify which category comes next. It specifies who will have legal standing when it arrives — and who will not.
We are watching the architecture being built. The timestamp matters.
*— Martin Niemöller died in 1984. He spent the last decades of his life insisting his poem was not about other people.*
Sources
Section I
– Tom’s Hardware / Nikkei Asia, Q1 2026 tech layoff aggregate, April 8: https://www.tomshardware.com/tech-industry/tech-industry-lays-off-nearly-80-000-employees-in-the-first-quarter-of-2026-almost-50-percent-of-affected-positions-cut-due-to-ai
– Challenger, Gray & Christmas via Bloomberg, April 2: https://www.bloomberg.com/news/articles/2026-04-02/us-job-cut-announcements-in-tech-keep-rising-with-ai-adoption
– LayoffAlert.org WARN Act tracker, April 2: https://layoffalert.org/layoffs-2026
– Oracle SEC filing / CNBC, March 31–April 1: https://www.cnbc.com/2026/03/31/oracle-layoffs-ai-spending.html
– AWS / Data Center Dynamics, April 10: https://www.datacenterdynamics.com/en/news/aws-scales-up-investment-commitment-for-mississippi-data-centers-to-25bn
– Prologis / Data Center Dynamics, April 10: https://www.datacenterdynamics.com/en/news/prologis-900mw-project-sail-gets-the-go-ahead-in-coweta-county-georgia
– Blackstone / Data Center Dynamics, April 10: https://www.datacenterdynamics.com/en/news/blackstone-acquires-minority-stake-in-data-center-firm-rowan
Section II
– Executive Order 14398, March 26: https://www.mondaq.com/unitedstates/government-contracts-procurement-ppp/1769956/new-executive-order-dei-practices-by-federal-contractors
– GSA SAM.gov proposal / Inside Higher Ed, April 1: https://www.insidehighered.com/news/diversity/2026/04/01/higher-ed-denounces-gsas-proposed-federal-funding-strings
– Trump AI EO preempting state laws, December 2025: https://www.credo.ai/blog/latest-ai-regulations-update-what-enterprises-need-to-know
– DOL / NSF TechAccess: AI-Ready America, April 7: https://pam.int/weekly-digest-on-ai-and-emerging-technologies-7-april-2026/
Section III
– Jennifer Harris, New York Times, April 8, 2026: https://www.nytimes.com/2026/04/08/opinion/ai-wealth-inequality-jobs-investment.html
– Thomas Friedman, New York Times, April 7, 2026: https://www.nytimes.com/2026/04/07/opinion/anthropic-ai-claude-mythos.html
– Anthropic Claude Mythos system card and Project Glasswing briefings, April 7–8, 2026
Notes from the Field publishes fortnightly. Out-of-cycle posts appear when the pattern demands it.
Leave a Reply