Regulating AI now follows three very different routes, and a news explainer carried on Wednesday 7 October 2026 by The Economic Times, The Edge Malaysia and Ahram Online set them side by side. The United States relies on voluntary pledges and executive orders. China layers detailed state rules on top of tight control of content. The European Union has a single rulebook, the AI Act, plus a growing set of online-safety laws that reach chatbots.
The timing matters. Calls to rein in the most powerful AI systems have grown louder since September, when the heads of several frontier labs asked in public to slow down. We covered that moment in our report on pacing the frontier. Each government has now answered in its own way, and the answers barely overlap.
This article sets out what the explainer reported about regulating AI, adds the dates, instruments and penalties behind each path, compares them in one table, and shows where they meet. It closes with a practical plan for UK and European organisations that build or buy AI and sell into more than one of these markets.
Table of contents
- Three Paths to Regulating AI, as the Explainer Laid Them Out
- The US Path: Regulating AI by Pledge and Executive Order
- China’s Path: Regulating AI Through State Control
- The EU Path: Regulating AI With a Rulebook
- Regulating AI Side by Side: The Comparison Table
- How Fast Each Path Bites
- Where Regulating AI Converges: Children and Frontier Security
- What Regulating AI Three Ways Means for UK Businesses
- A Practical Plan for Firms Selling Into All Three Markets
- What to Watch Next in Regulating AI
- Regulating AI: Frequently Asked Questions
- References
Three Paths to Regulating AI, as the Explainer Laid Them Out
The explainer’s summary is short and blunt. The US, China and the EU “are taking different paths to govern artificial intelligence, as calls grow worldwide to rein in its most powerful systems.” The same text ran in The Economic Times at 03:50 UTC, and in The Edge Malaysia and Ahram Online within hours.
The one-line version of each path
In the US, the explainer says President Donald Trump “has built his AI policy almost entirely from voluntary industry pledges and executive orders”. In China, the government “already has a complex, evolving set of rules on how AI is used”, and it censors AI-generated content much as it does other online information. In Europe, the AI Act is the centrepiece, which Brussels calls the world’s most comprehensive law on the technology.
Why the differences are not just style
These are not three dialects of one idea. They rest on different answers to a basic question about regulating AI: who should hold the brake? Washington leaves it largely with the companies. Beijing keeps it firmly with the state. Brussels writes it into law and gives regulators the power to fine. Those choices decide who a business answers to when something goes wrong.
The US Path: Regulating AI by Pledge and Executive Order
The American approach to regulating AI is built from orders the president can sign and pledges the companies can make. Congress, which the explainer describes as “politically divided and heavily lobbied by Big Tech”, has passed no federal AI law. It is out of session until after the midterm elections.
From Biden’s order to an AI boom
On his first day back in office, Trump scrapped Joe Biden’s 2023 order that required developers of the most powerful AI systems to share safety test results with the government. In July 2025 the White House published an AI Action Plan listing more than 90 measures to speed up development. A companion order, “Preventing Woke AI in the Federal Government”, requires agencies to buy only models deemed ideologically neutral.
Security moved to the front in 2026
The explainer says the release of AI models capable of powerful cyberattacks shifted the focus this year. A June order set up a voluntary system under which developers can give the government up to 30 days of early access to their most powerful models. It is a test-before-release idea, but it stays voluntary, and it carries no licence or approval step.
The September accord and the rename
On 29 September the heads of major AI companies signed a White House accord pledging independent audits and board-level oversight of cutting-edge models. Trump called it “morally binding”, and the explainer notes “the pact carries no penalties”. We explained how its four layers work in our guide to the AI safety accord. The same day, Trump ordered agencies to say “super intelligence” instead of “artificial intelligence”.
The US timeline at a glance
The table lists the main US instruments in date order, with the explainer or our own earlier reporting as the source. None of them creates a duty that an AI company can be fined for ignoring.
| Date | Instrument | What it does | Binding on AI companies? |
|---|---|---|---|
| 20 Jan 2025 | Revocation of Biden’s 2023 order | Ends duty to share safety test results | Removes a duty |
| July 2025 | AI Action Plan | More than 90 measures to speed up development | No |
| July 2025 | “Preventing Woke AI” order | Agencies buy only models deemed neutral | Only as a sales condition |
| June 2026 | Early-access order | Up to 30 days of pre-release access | No, voluntary |
| 29 Sep 2026 | White House accord | Audits and board oversight | No penalties |
| 29 Sep 2026 | Terminology order | “Super intelligence” in official use | Agencies only |
| 4 Oct 2026 | Super Intelligence Force | White House body chaired by Jay Clayton | No |
The states are filling the gap
The federal picture is not the whole story. The explainer reports that Trump pushed to block state AI laws, but Congress “repeatedly refused to override them”, and state laws are taking effect nationwide. For a business, that means regulating AI in the US is increasingly a state-by-state matter, even while Washington prefers pledges. Our report on voluntary AI safeguards explains why critics say existing law only acts after the harm.
China's Path: Regulating AI Through State Control
China’s route to regulating AI is the most detailed on paper and the most centralised in practice. The explainer describes “a complex, evolving set of rules on how AI is used, including by children”, sitting beside the same censorship regime that governs other online content.
The companion-bot rules that took effect in July
The newest layer covers AI companions, one of the fastest-growing uses of generative AI. The Interim Measures for the Administration of AI Anthropomorphic Interaction Services were issued on 10 April 2026 by the Cyberspace Administration of China and four partner agencies, and took effect on 15 July 2026. The explainer says they aim to curb emotional dependency and ban digital clones of real people made without their consent.
What the companion rules require
According to Just Security’s summary, providers cannot offer minors virtual partners or relatives, and anyone offering other anthropomorphic services to a child under 14 needs parental consent. Services must not be designed to induce emotional dependence or replace real relationships. Penalties escalate from warnings to service suspension and fines, and providers also file their algorithms under China’s 2022 rules.
A stack of earlier rules
The companion measures sit on top of rules China has built since 2022. Each targets one use of the technology rather than AI in general, which is a different way of regulating AI from the EU’s single horizontal act.
| In force from | Chinese rule | Main target |
|---|---|---|
| 1 Mar 2022 | Algorithmic recommendation provisions | Feeds, ranking and algorithm filing |
| 10 Jan 2023 | Deep synthesis provisions | Deepfakes and synthetic media |
| 15 Aug 2023 | Interim measures for generative AI services | Public chatbots and image generators |
| 1 Sep 2025 | Labelling measures for AI-generated content | Visible and hidden labels |
| 15 Jul 2026 | Anthropomorphic interaction measures | AI companions and minors |
Control, not a pause
China’s leaders talk openly about risk. The explainer quotes President Xi Jinping saying this summer that AI should be “always under human control”, with more “laws and regulations, technological monitoring, early warning, and emergency response systems”. Premier Li Qiang has warned that the world risks “losing control” of frontier technologies if governments are too slow to regulate them.
Why Beijing rejects a slowdown
Yet a pause is not on the table. Oxford China Policy Lab analysts wrote in September that “a general pause or deliberate slowdown in AI development remains outside the mainstream Chinese policy posture.” Many in China see US export controls as a slowdown already imposed on them. Our coverage of the AI safety debate records Beijing’s response to the calls from US labs.
Security as the lens
Last month China’s Minister of State Security, Chen Yixin, warned that “hostile forces” are abusing generative AI to “fabricate political rumors and spread harmful information.” That framing explains a lot. For Beijing, regulating AI is part of national security and information control first, and consumer protection second.
The EU Path: Regulating AI With a Rulebook
Europe chose law over pledges when regulating AI. The AI Act was adopted in 2024 and entered into force on 1 August 2024, with its duties phased in over several years. Brussels describes it as the world’s most comprehensive set of rules on the technology.
A rulebook that was then delayed
The explainer notes that European companies, fearing the rules would stifle innovation, “later won a delay on rules covering high-risk systems”. The vehicle was the Digital Omnibus on AI, provisionally agreed on 7 May 2026, which pushed most high-risk duties to 2 December 2027. Bans on prohibited practices have applied since 2 February 2025, and rules for general-purpose models since 2 August 2025.
Penalties with real teeth
Unlike the US accord, the AI Act names its fines. Breaching a prohibition can cost up to €35 million or 7% of worldwide annual turnover, whichever is higher. Most other breaches reach €15 million or 3%, and supplying incorrect information to regulators up to €7.5 million or 1%. Our EU AI Act compliance checklist maps those duties for UK firms.
Bars are scaled to the largest ceiling, 7%, so each width is the percentage divided by 7: 6 divided by 7 is 85.7%, 3 divided by 7 is 42.9% and 1 divided by 7 is 14.3%. The US accord names no penalty, so its bar is a sliver. China’s companion rules escalate to fines but publish no turnover percentage, so they are left off.
Pacing the frontier, the European way
Von der Leyen used her State of the Union speech on 16 September to back the slowdown idea. “Models being developed will allow hacking on a level we never thought possible,” she said, according to Euronews. She said the AI rules put “Europe in the position to shape global efforts”, and that she would invite the main frontier labs to discuss how to support efforts “to pace the frontier.”
Industry is less keen
European firms have been warier. The explainer reports they fear a slowdown “could be an attempt by US giants to lock in their dominance”. Mistral, Europe’s best-funded lab, launched a trillion-parameter model the day before the explainer ran. That tension, between safety rules and catching up, runs through every European debate about regulating AI.
Other EU tools: platforms and children
The AI Act is not Brussels’ only lever. On 31 August the Commission designated ChatGPT a very large online search engine under the Digital Services Act, the first AI chatbot on that list, as we noted in our piece on chatbot advertising. On 17 September it proposed the KIDS Act, which would stop AI companions from simulating relationships “in ways that create emotional dependency” in children.
Regulating AI Side by Side: The Comparison Table
Put the three paths to regulating AI in one grid and the gaps become obvious. This table uses the explainer’s account plus the primary instruments named above. “Binding” means a company can be penalised for ignoring the rule.
| Question | United States | China | European Union |
|---|---|---|---|
| Main tool | Executive orders, pledges | Targeted state rules | AI Act plus DSA |
| Federal or national AI law | None | Several targeted measures | One horizontal act |
| Frontier model testing | Voluntary, up to 30 days | Filing and state review | General-purpose model duties |
| Named penalties | None in the accord | Warnings to suspension and fines | Up to 7% of turnover |
| AI companions and children | State laws vary | In force since 15 Jul 2026 | KIDS Act proposed 17 Sep 2026 |
| Position on a slowdown | Pledge, no brake | Rejects a pause | Backs “pacing the frontier” |
| Who holds the brake | Companies | The state | Law and regulators |
Reading the grid
Three patterns in regulating AI stand out. The US has the fewest binding rules and the most announcements. China has the most rules aimed at specific uses, enforced by agencies with wide discretion. The EU has the clearest penalties but the slowest timetable, because a law takes years to phase in and can be amended on the way.
How Fast Each Path Bites
Speed is a hidden variable in regulating AI. A pledge applies the day it is signed but cannot be enforced. A law can be enforced but may take years to apply. The chart counts the months between adoption and the date each rule started to apply.
Bars are scaled to the longest wait, 40 months, so each width is the month count divided by 40: 12 divided by 40 is 30%, 6 divided by 40 is 15% and 3 divided by 40 is 7.5%. China’s figure rounds 96 days to three months. The accord applied the day it was signed, which is why it shows zero.
What speed tells a buyer
China moved from publication to enforcement in about three months. The EU’s slowest duties arrive more than three years after the law did. The US accord is instant and unenforceable. A supplier’s promises therefore age at different rates in each market, and contracts should say which rules the supplier commits to meet, by when.
Where Regulating AI Converges: Children and Frontier Security
For all the differences, two themes now appear on every path to regulating AI. Both are worth watching, because they are where a single global practice is most likely to emerge.
Children and AI companions
China’s companion rules, the EU’s KIDS Act proposal and a wave of US state laws all target emotionally manipulative chatbots and minors. The explainer links the EU proposal to “a spate of suicides worldwide linked to children’s interactions with such assistants”. Any product with a companion or persona feature should expect age checks and dependency safeguards wherever it is sold.
Frontier security testing
The second theme in regulating AI is cybersecurity risk from the most capable models. The US asks for up to 30 days of voluntary early access. The EU wants model evaluation and early warning with partners such as Canada and the UK. China speaks of monitoring and emergency response. The methods differ, but all three now treat frontier models as a security question.
The words differ, the worry does not
Even the vocabulary is converging on control. Xi talks of keeping AI “under human control”. Von der Leyen talks of pacing the frontier. The US accord promises audits and board oversight. Regulating AI is becoming, in all three capitals, a debate about who can stop a model, not whether one should ever be stopped.
What Regulating AI Three Ways Means for UK Businesses
Few UK organisations build frontier models, but many buy AI from US vendors, sell into the EU, or source components and services that touch China. Each of those links imports a different regime for regulating AI.
If you sell into the EU
The AI Act applies to providers that place systems on the EU market, wherever they are based. A UK firm selling an AI product in Europe therefore faces the same duties and fines as a European one. The high-risk delay to December 2027 buys time, not an exemption.
If you buy from US vendors
Your supplier’s main safety commitment may be a voluntary pledge. That is not a reason to avoid US tools, but it is a reason to put the protections into your own contract: audit rights, incident notice, data location and the right to switch. Our analysis of AI self-regulation lists the gaps the pledge leaves open.
If China is in your supply chain
Products built for Chinese users must meet Chinese content and labelling rules, and companion features face the July measures. Data and model exports raise separate questions. Treat China as its own compliance track, not a variant of the EU one.
A Practical Plan for Firms Selling Into All Three Markets
The safest way to cope with regulating AI three ways is to build to the strictest common denominator, then add local layers. The table turns that into actions, owners and the regime each one answers.
| Action | Main driver | Owner |
|---|---|---|
| Inventory every AI system and its markets | EU AI Act scope | Product and legal |
| Classify risk level per system | EU high-risk categories | Compliance |
| Label AI-generated content | China labelling, EU transparency | Product |
| Age checks and dependency safeguards on companion features | China measures, KIDS Act | Product and safety |
| Write audit and incident clauses into vendor contracts | US reliance on pledges | Procurement |
| Track state laws where you sell in the US | US state legislation | Legal |
| Review the plan every quarter | All three | Board sponsor |
Start with the inventory
Nothing in regulating AI works without a list of the AI systems you use or sell, what they do and where their users are. Most firms find more than they expected, because AI features arrive inside ordinary software updates. The inventory is also the document a regulator in any of the three blocs will ask for first.
Build once, localise the edges
Content labelling, logging, human oversight and incident response are wanted almost everywhere. Build them once to the EU standard. Then add the local layers: Chinese filing and content rules, US state duties and the EU’s high-risk documentation. That keeps regulating AI from turning into three separate programmes.
Keep a board-level owner
All three regimes now point at the top of the organisation, whether through EU accountability, the US accord’s board committees or China’s summons of legal representatives. A named board sponsor makes sure that changes in any one market are noticed and funded.
What to Watch Next in Regulating AI
The three paths to regulating AI will not stand still. Several dates in the coming months will show whether they drift further apart or start to meet.
In the United States
Watch Congress after the midterms, the work of the new Super Intelligence Force, and whether any company is tested against its accord pledges. Our explainer on the Super Intelligence Force covers its 120-day brief.
In China and the EU
In China, watch the first enforcement cases under the companion rules and any new measures for agents. In the EU, watch the KIDS Act in the European Parliament and Council, the frontier-lab talks von der Leyen promised, and the run-up to the December 2027 high-risk deadline.
Regulating AI: Frequently Asked Questions
What are the three paths to regulating AI?
The US relies on voluntary pledges and executive orders. China uses targeted state rules plus tight content control. The EU uses the AI Act, a single law with named fines, plus online-safety laws such as the Digital Services Act.
Does the US have a federal AI law?
No. The explainer reports that Congress has yet to pass one. US policy is made through executive orders, a voluntary accord signed on 29 September 2026, and a growing number of state laws.
What do China’s AI companion rules require?
They took effect on 15 July 2026. They bar virtual partners or relatives for minors, require parental consent for users under 14, and forbid designs that induce emotional dependence. The explainer adds that they ban digital clones of real people made without consent.
When do the EU AI Act’s high-risk rules apply?
After the Digital Omnibus delay, most high-risk duties apply from 2 December 2027. Prohibited practices have been banned since 2 February 2025, and general-purpose model rules have applied since 2 August 2025.
Which path to regulating AI is strictest for a business?
On paper, the EU, because its fines reach 7% of worldwide turnover. In practice China’s rules can be enforced faster and with more discretion. The US is lightest at federal level, but state laws are tightening.
References
US, China, EU: Three paths to regulating AI (The Economic Times)
US, China, EU: three paths to regulating AI (The Economic Times on MSN)
What to Know About China’s First AI Companion Rules (Just Security)
European Commission Proposes EU KIDS Act (Davis Wright Tremaine)
EU’s von der Leyen calls for pacing frontier AI models (Euronews)
EU AI Act Article 99: Penalties (artificialintelligenceact.eu)
EU AI Act implementation timeline (artificialintelligenceact.eu)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
