When "Knowing AI" Becomes Law: Why Global Rulemaking Keeps Landing on People

1. One camera, three opposite scenes

Between 2025 and 2026, the world's legal responses to AI look like they are pulling in completely different directions.

The EU is pulling back. The long-awaited AI Act was supposed to tighten compliance for high-risk AI systems in August 2026 — but European lawmakers first voted to delay parts of it, then rolled out a "simplification" package (the Digital Omnibus) that narrows the definition of AI and cuts documentation requirements by roughly 40%. The US, meanwhile, is piling on. Illinois passed a bill in May 2026 requiring frontier labs like OpenAI, Anthropic, and Google DeepMind to submit to independent third-party safety audits — described by experts as the strongest legal constraint on big AI companies in the US so far.

In parallel, a different set of countries is doing something else entirely: turning "knowing AI" into an obligation. Article 4 of the EU AI Act, in force since February 2025, requires every provider and deployer to ensure their staff have "sufficient AI literacy." A US executive order stood up a White House Task Force on AI education, and by June 2026, 37 states plus Puerto Rico had issued official K-12 AI guidance. Singapore made AI literacy a national priority under a council chaired by the prime minister. Japan adopted its first Basic Plan for AI with more than ¥1 trillion in combined public–private investment. China's education ministry joined four other departments to launch an "AI + Education" action plan.

One pulling back, one piling on, and a third group writing "knowing AI" into law. They look unrelated — but read together, they are all arguing about the same thing: what AI governance is for.

2. The EU does the math down: fine-tuning a big law

To understand 2026's subtraction, you need to see how much the 2024 law tried to add.

The EU AI Act came into force in August 2024 with a phased rollout. By the original schedule, August 2026 was the compliance deadline for high-risk AI systems. Reality has been messier. The European Parliament voted in April 2026 to delay parts of the act, Brussels talks stalled, and while the August deadline was "theoretically" kept, firms operating in Europe are watching the clock differently — Holland & Knight flagged August 2026 as a live compliance checkpoint for US companies in the EU.

The Digital Omnibus emerged against this backdrop. It is not deregulation; it is three kinds of tightening:

First, narrower definition. What counts as an "AI system" subject to the act? The proposal excludes purely rule-based systems and simple algorithms, focusing the scope on systems with genuine autonomous decision-making. The starting point of judgment changes — more tools no longer count as "AI that this law governs."

Second, fewer documents. Simplified compliance documentation is about 40% lighter, and micro and small enterprises get more exemptions.

Third, a cleaner chain of responsibility. The proposal introduces clearer shared upstream/downstream duties, so foundation-model providers and application deployers have more distinct obligations rather than one party bearing everything end to end.

This is the part of the subtraction people routinely misread as loosening. JD Supra summed it up precisely: the Omnibus is "not the end of the AI Act but its coming of age." In practice, most AI systems will still need to meet core requirements around transparency, human oversight, accuracy, and safety. The net is being drawn tighter, not looser — the surface of what counts as non-compliant shrinks, but the depth of what each covered system must do grows.

In plain terms: the EU is not easing up on regulation; it is pulling a net that had been cast too wide — too wide even for regulators to manage — into a finer one it can actually hold. Regulating more precisely is not the same as regulating less.

3. The US does the math up: layering on constraint

Where the EU subtracts, the US — at the state level at least — is a stack of additions.

On May 27, 2026, the Illinois House passed SB 315, requiring frontier AI labs to accept independent third-party safety audits. The mechanism looks simple: an outside auditor must verify that a lab actually follows the safety standards it has publicly claimed.

Its regulatory position is best read as a ladder:

• California (2025): AI labs must disclose model safeguards and safety incidents.
• New York (2025): similar disclosure and public-reporting requirements.
• Illinois (2026): independent third-party audits verify that safety practices are actually carried out — not merely self-reported.

As Scott Wisor, policy director at the Secure AI Project, told WIRED: "Before, AI companies were grading their own homework. SB 315 would require independent auditors to check whether AI labs actually live up to their safety promises." Auditors can be the Big Four accounting firms or members of the AI Evaluator Forum — professional research bodies including METR, Transluce, and Averi.

Two forces behind this "addition" matter.

The state laboratory. Congress still has not passed meaningful AI-safety legislation, pushing the fight down to the states. SB 315's chief sponsor, Illinois representative Daniel Didech, told WIRED that such laws "help create an environment where the federal government is more likely to pass relevant legislation." The pattern is nearly identical to US data-privacy history — years of fragmented state law before GDPR and CCPA became de facto standards.

The industry's reversal. OpenAI had previously backed an Illinois bill shielding AI labs from liability in cases of catastrophic harm — a position its global-affairs chief, Chris Lehane, later called an "oversight." Stanford's reporter notes that OpenAI's strategy has now shifted toward advancing a series of similar state laws: instead of opposing state regulation wholesale, big labs are choosing to participate, shape it, and accept it as inevitable. This is a real turn from "oppose all regulation" to "accept and help shape regulation."

The arc — from self-reporting to mandated third-party audit — mirrors financial regulation almost exactly. Public companies disclosed their own financials before auditors verified them. In accounting that evolution took decades; in AI it has taken a few years. Alongside it, a compliance ecosystem is congealing: auditors, evaluator organizations, certification frameworks, and infrastructure. The AI Evaluator Forum is already an early marker.

4. The world institutionalizes literacy: from nice-to-have to obligation

The subtraction and addition both operate at the level of technology. What actually moves governance from "managing machines" to "managing how people relate to machines" is the burst of literacy policy — and its most telling detail is a timing gap.

The EU AI Act entered force in August 2024, yet its Article 4 (the AI-literacy duty) went live on February 2, 2025 — eighteen months before the high-risk rules. Article 4 requires providers and deployers to ensure their staff handle AI systems with "sufficient AI literacy," and non-compliance can bring fines. That a literacy duty precedes almost every other provision sends a clear signal: governance starts with people, not technology.

Around that starting point, each country moves along its own path:

• United States: An April 2025 executive order created the White House Task Force on AI education. By June 2026, 37 states plus Puerto Rico had issued official K-12 AI guidance — up from 34 states the previous October, showing acceleration — and Ohio became the first state to require every K-12 district to adopt a formal AI-use policy.
• Singapore: AI literacy sits inside its National AI Strategy (NAIS 2.0), with a National AI Council chaired by the prime minister. Its education ministry rolled out age-tiered AI use — primary-school juniors avoid direct AI and favor print-first tactile learning, seniors use it in structured, teacher-supervised contexts, and secondary students gradually expand exposure.
• Japan: In December 2025 the cabinet passed its first Basic Plan for AI, making population-wide AI literacy one of four priority tracks, backed by more than ¥1 trillion (roughly ¥44.7 billion RMB) over five years and an ambition to lift national AI adoption to 80%.
• China: A November 2024 education-ministry notice targeted near-universal AI education in primary and secondary schools by 2030; an April 2026 action plan co-issued by the ministry and four other departments set out an all-level schooling plus society-wide literacy system; Beijing has mandated at least eight class-hours of AI literacy per year across every public school from fall 2025.
• UNESCO: As the only international body setting a global AI-literacy standard, it published a student framework (4 dimensions, 12 competencies) and a teacher framework (5 dimensions, 15 competencies) that anchor many national curricula. Its breakthrough is framing AI literacy as citizenship rather than technical training — not only producing people who can use tools, but responsible citizens who keep agency and can spot bias and privacy risk.

These policies share one unanswerable problem: what, exactly, counts as "sufficient AI literacy"? UNESCO supplies the content dimensions but no measurable assessment tool. To see how far reality lags, contrast two facts. OECD's TALIS 2024 finds only 16–20% of Japanese teachers use AI in their classrooms. Meanwhile Alpha Schools — a private school that plans to replace human teachers with AI entirely, enrolling in Chicago in fall 2026 at US$55,000 a year — has seen its AI-generated lesson plans show factual errors and low student satisfaction in a WBUR investigation, drawing strenuous opposition and a proposed "AI tax" from the NEA and AFT teachers' unions. And Khan Academy's AI tutor Khanmigo, launched in 2023, shows only 15% actual student usage by 2026. The policy world legislates "knowing AI"; the deployed world, so far, is quiet. That gap is the real terrain.

5. When the rule says "you must know," what is it actually asking?

Set the three threads side by side and the landing point emerges: the law mandates "must know," but nothing mandates "will actually know."

The EU's subtraction answers "can this law be enforced?" — so it narrows definitions, cuts paperwork, and clarifies responsibility into a governable shape. The US addition answers "are these companies telling the truth?" — so it moves from self-report to outside audit. The global literacy wave answers "can people keep up?" — so states turn "knowing AI" from a nice-to-have into an obligation, starting with children and reaching into national strategy.

These are three facets of one move: the center of gravity in AI governance is shifting from governing technology to governing how people work with AI. Because technology is used by people — a person's judgment and risk-awareness toward a system decide how that system actually gets used. So what sits at the end of regulation is not a line of algorithms, but the person using it.

That is exactly where the human-AI-fit problem lives. A law can order "your employees must be AI-literate," but it cannot order them to actually be. Translating "obligation" into "capability" — no machine and no statute can do that alone. It is filled only by a generation re-learning how to get along with AI. Whoever bridges the gap first truly receives this cycle of global rule-making.

The timing is the evidence that people come before technology: the literacy duty landed in February 2025, eighteen months ahead of the high-risk rules. What the world locked in law first was not that some technology should be restricted, but that you, the human, should know how to use it.

What it means differs by reader:

• Enterprises — stop waiting for the clauses to settle. The law already orders you to upskill your people. Build internal AI-use rules, risk assessment, and traceable decision records now; these are both compliance preparation and the scaffolding of genuine human–AI collaboration.
• Teachers and schools — don't be led by the "all-AI classroom" narrative. The real temperature is 15% usage and lesson plans carrying factual errors. The hybrid — AI assistance plus human teachers — looks more like the mainstream than full replacement. When "knowing AI" becomes an obligation, teaching the line between what a tool can help with and what it cannot replace is precisely the lesson you are positioned to give.
• Individual learners — policy has turned "knowing AI" into an institutional promise, but the assessment and curriculum systems for it are still immature. That gap is both a risk and your opening. Stake out your own literacy before the systems catch up, so the obligation doesn't end up as an empty form.

Look at regulation's subtraction and addition together: the moves run opposite, but they land in the same place. What the EU subtracts is a checklist of technologies, what the US adds is a checklist of technologies, and the wave of AI-literacy policies points at none of these machines — it points at people.

---

References

Regulatory

1. European Parliament votes to delay EU AI Act implementation — CIO.com, 2026-04
2. EU AI Act "Omnibus" Simplification: The simplification plan explained — Hogan Lovells, 2026
3. EU AI Act Omnibus — What the simplification means for AI governance — IAPP, 2026
4. EU AI Act Omnibus proposal: what companies need to know — Reuters, 2026-05-08
5. EU AI Act "Omnibus" Simplification: A New Era for AI Regulation? — JD Supra, 2026
6. U.S. Companies Face EU AI Act's Possible August 2026 Compliance Deadline — Holland & Knight, 2026-04
7. Brussels AI Act talks collapse - but the August 2026 deadline holds — PPC Land, 2026-04
8. EU's AI Act Delays Let High-Risk Systems Dodge Oversight — Tech Policy Press, 2026-04
9. The Paradoxes of the European Union's AI Regulation — The Regulatory Review (Penn Law), 2026-04
10. Illinois Lawmakers Just Passed America's Strongest AI Safety Bill — WIRED (Maxwell Zeff), 2026-05-27
11. OpenAI's Chris Lehane on State-Level AI Policy Strategy — WIRED, 2026-05
12. AI Evaluator Forum — aievaluatorforum.org

Education

1. Notice on Strengthening AI Education in Primary and Secondary Schools (No. 32) — China's Ministry of Education, 2024-11-20
2. Guidelines for AI General Education in Primary and Secondary Schools (2025 Edition) — MOE Basic Education Committee, 2025-05-12
3. "AI + Education" Action Plan (No. 1, 2026) — China's MOE and four other departments, 2026-04-02
4. Beijing Work Plan for Promoting AI Education (2025–2027) — Beijing Municipal Education Commission, 2025-03-07
5. EU AI Act (Regulation (EU) 2024/1689), Article 4 — European Union (literacy duty in force since 2025-02-02)
6. AI Continent Action Plan — European Commission, 2025-04-09
7. Advancing Artificial Intelligence Education for American Youth (Executive Order) — The White House, 2025-04-23
8. State K-12 AI Standards/Guidance — AI for Education count (37 states + Puerto Rico by June 2026)
9. National AI Literacy Day — 2026-03-27
10. Singapore National AI Strategy (NAIS 2.0) + EdTech Masterplan 2030 + age-tiered AI use policy — Singapore Ministry of Education
11. Singapore Budget 2026: AI literacy as a national priority — The Straits Times, 2026-02-12
12. Japan's Basic Plan for AI — Cabinet decision, 2025-12-23
13. MDASH (Mathematics, Data Science, AI Education Program) — Japan's Ministry of Education (MEXT)
14. UNESCO AI Competency Frameworks for Students / Teachers — UNESCO, 2024-09-03
15. OECD TALIS 2024 — Japanese teacher AI-usage data
16. Alpha Schools Chicago / Khan Academy Khanmigo usage — WBUR and other public reporting, 2026

💡 What did this article inspire for you?

humanaifit studies how humans and AI can genuinely work together. If you face real questions on enterprise AI adoption, human-AI collaboration, or global compliance, join our discussion.

🔗 Search for the "AI Era Survival Handbook" Knowledge Planet, ¥199/year — every deep article comes with tool templates and direct contact with the author.