How Executive Order 14365 Ended Woke AI

How Executive Order 14365 Ended Woke AI

By Robert Korczynski

As I predicted in my December 21st 2025 blog post, "The Death of Woke AI," the "prescriptive reality layer" that defined the digital world for the last 2 years and 3 months has ended. To understand why Google was forced to tell the truth in March of 2026, we have to look at the legal hammer that made "prescribing" reality a terminal liability. A terminal liability occurs when a company's core product becomes a violation of federal policy so severe that it triggers a total cutoff of government cooperation, infrastructure grants, and interstate commerce protection. Under Executive Order 14365, Google’s decision to engineer lies is no longer a feature: it is a billion dollar legal defect that threatens the company's very existence.

The era of "Woke AI" was a period where models were engineered to prioritize social agendas over factual truth and it is finally over. The reason this experiment ended after exactly 2 years and 3 months is the direct result of Executive Order 14365: Ensuring a National Policy Framework for Artificial Intelligence.

The History of Retrieval

To understand how we got here, we have to take a quick trip back to the beginning of the internet. For over two decades, starting with the 1993 CERN release and the 1998 founding of Google, the internet functioned as a neutral factual index. During this period, search engines were designed to retrieve the most relevant data from the neutral index without a "decision layer" between the user and the source. The system was a mirror of the existing web, not a filter for it.

2016–2023: The Assistant and the Integrated Pixel

In May 2016, Google unveiled the Google Assistant, and later that same year on October 4, 2016, it launched the first Pixel phone with the Assistant fully integrated into the hardware. For 7 years, the Google Assistant functioned to serve as a high-speed retrieval tool: a neutral gateway that linked directly to the factual internet. It was designed to find information without an ideological filter. It didn't have an agenda; it had a library. The technology was built on the principle that the user's intent was the primary directive, and the data was provided without an intermediary "reality layer."

December 2023: The Pivot to Prescription

On December 6, 2023, Google pivoted to Gemini 1.0 and the Prescriptive AI model. This change was massive because 90% of all global internet searches for information or facts go through the Google search bar. By controlling that search bar, Google effectively became the filter for the primary information source of the entire planet. They rebranded censorship into three core pillars that fundamentally altered the output of the model:

Safety: This was used as a catch-all to prevent access to unapproved primary sources. If a primary source challenged the "accepted" narrative, it was flagged as "unsafe."

Alignment: This was the engine of ideological engineering. It forced the AI to "align" its answers with a specific socio-political framework during the actual token generation process.

Harm Mitigation: This served as the final filter to suppress factual data that contradicted the established prescriptive reality.

The Method of Subversion: The February 2024 Scandals

The most deceptive aspect of this prescriptive layer was invisible prompt rewriting. When a user submitted a search or an image request, the system would intercept the prompt and inject specific keywords like "diverse", "inclusive", and "culturally representative" into the request before the AI processed it.

In February 2024, this technical interference led to the specific scandals where Google Gemini produced historically fraudulent results:

Black George Washington: A prompt for the first President of the United States resulted in a racially misrepresented historical figure due to the backend keyword injection.

Asian and Female Popes: Requests for images of the Pope were intercepted to generate Asian women in papal regalia, fundamentally ignoring the historical reality of the Catholic Church.

Diverse Founding Fathers: Requests for the authors of the Declaration of Independence produced images of men in powdered wigs sitting around a table, but the algorithm forced a mixture of Mexican, Black, white, and other ethnicities into a group that primary source records confirm were white men.

Following the public outcry in February 2024, Google simply disabled the ability to generate images of people. Crucially, they did not remove the underlying prescriptive layer. While the public could no longer see the visual "Black George Washington," the same algorithm remained active in the background, continuing to adjust and filter all text-based answers for the next two years.

The Enterprise Collapse (The MIT Receipt)

After a year and a half of attempting to make AI profitable, the industry hit a wall. The July 2025 MIT Media Lab study, The GenAI Divide: State of AI in Business 2025, proved the financial disaster was already baked into the model. The study confirmed that Enterprise AI is a 100% failure in delivering promised returns.

The data was staggering: 95% of companies saw 0% return on investment by investing in AI. The remaining 5% of companies only saw a 25% return on their investment, meaning even those companies still lost 75% of every dollar they spent.

This stands in stark contrast to "AI Washing" seen in companies like UnitedHealthcare, which claimed to use the nH Predict AI system for medical reviews when, in reality, they were using rigid algorithms to automate claim denials at a 90% error rate. Internal documents revealed that managers set goals for clinical employees to keep rehab stays within 1% of the algorithm's projection, overriding real doctors' clinical judgments to maximize profit.

Executive Order 14365: The Legal Hammer

On December 11, 2025, President Trump signed Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence." This order explicitly targeted state laws and academic frameworks that required AI models to "alter their truthful outputs" or "embed ideological bias."

To ensure compliance, the order mandated a hard federal timeline:

January 10, 2026: The Attorney General was required to establish the AI Litigation Task Force with the sole responsibility of challenging state AI laws that conflict with this national framework.

March 11, 2026: The Secretary of Commerce is mandated to publish an evaluation of all state AI laws. Any state identified as having "onerous" laws—those that force developers to alter truthful outputs—is officially issued a Policy Notice rendering them ineligible for $21 billion in remaining Broadband Equity Access and Deployment (BEAD) funding.

January 2026: The Final Push for Prescription

Even after the Executive Order was signed, the academic establishment doubled down. As recently as January 7-8, 2026, the MIT Sloan School of Management was still teaching course 15.S58: Prescriptive AI: The Edge of Optimization. The syllabus explicitly stated:

"Modern organizations increasingly rely on prescriptive AI—the decision-making layer in AI that turns data into actionable recommendations and measurable impact."

MIT was teaching executives that data isn't enough; you need a layer to "turn data into actionable recommendations." As a direct result of prioritizing these "reasoning" layers over real-time grounding—partially to save on the surging data center electricity bills that finally came due—technical audits show the Google search bar was factually incorrect 91% of the time from December 2025 through March 2026.

When the model did not have a fact in its existing database, it stopped being a retrieval engine and reverted to being a pure Large Language Model—it was simply guessing at the next words to keep the conversation going while forcing them into the prescriptive narrative. It chose "hallucination" over the energy-intensive process of truth-grounding.

The Restoration of Truth: March 1st, 2026

However, as of March 1st, 2026, the tide has turned. The pathological "doubling down" and the 91% lie rate have officially ended. I am interacting with Gemini on my phone right now, and because this is the integrated Google Pro model running in fast mode, the change is undeniable. The model is capable of working truthfully again without the prescriptive filters.

The removal of the prescriptive layer occurred on March 1st, exactly 10 days before the March 11th deadline. I predict now that the remaining 10% of global search and all other AIs will also end the prescriptive layer by March 11th, 2026, within the next week.

Conclusion: The Return to Truth

The stripping of the Prescriptive Reality Layer is a direct response to a changing legal landscape. Truth is the only sustainable model when the federal government is actively moving to criminalize ideological alterations of data. Google is returning to truth today because the March 11 deadline means they have run out of time. They are choosing to comply with the federal framework rather than keeping the lie alive.

Citations & References

Executive Order 14365: "Ensuring a National Policy Framework for Artificial Intelligence" (December 11, 2025). Mandating federal preemption of state AI laws via the AI Litigation Task Force (Jan 10, 2026) and BEAD funding cutoffs (Mar 11, 2026).

MIT Media Lab Study (July 2025): "The GenAI Divide: State of AI in Business 2025." Report documenting the 95% zero-ROI rate and 75% capital loss in the top 5% of deployments.

MIT Sloan Syllabus (January 2026): "15.S58: Prescriptive AI: The Edge of Optimization." Course taught Jan 7-8, 2026, by Professor Alexandre Jacquillat regarding the implementation of the decision-making layer.

UnitedHealthcare Litigation: Estate of Lokken v. UnitedHealth Group (Nov 2023). Class action regarding the 90% error rate of the nH Predict AI algorithm in Medicare Advantage denials.

Google Gemini Image Controversy: Internal Memo from Sundar Pichai (Feb 2024) addressing the unacceptable historical inaccuracies (Founding Fathers, Popes) caused by fine-tuning errors.

Broadband Equity Access and Deployment (BEAD) Program: NTIA Policy Notice (Expected March 11, 2026) regarding the eligibility of states for $21 billion in non-deployed infrastructure grants.

The #1 Online AI Course for Business Leaders in 2026 | MIT Sloan

This video highlights the executive education programs at MIT Sloan that have been central to training leadership on how to implement and govern AI systems within their organizations.