The UK CMA launches a “fact-finding” review of the AI market, including LLMs, to assess opportunities, guardrails, and principles to protect competition

CMA chief Sarah Cardell says the regulator will examine technology behind software such as ChatGPT

Financial Times 2023-05-04

Context & Ripple Effects

This was the CMA's opening move from broad AI-market scrutiny toward a competition framework for foundation models. By September, it had set out seven principles for foundation-model competition and planned engagement with major model, platform and chip companies.

The review matters as an early marker of the regulator treating AI's commercial stack—not just individual products—as a competition question. Later CMA work on an interconnected web of AI investments and deals and specific Microsoft and Google partnerships shows how that initial inquiry translated into transaction-level scrutiny.

First-order effects

The CMA begins collecting evidence on how LLM-related markets work, putting developers, platforms and other market participants on notice that competition safeguards will be assessed alongside AI's opportunities.
The regulator can use the review to define the guardrails and principles it considers relevant before deciding whether more targeted interventions are needed.

Second-order effects

Companies building, distributing or financing LLM services face pressure to explain how their partnerships and routes to market affect rivalry; this is consistent with the CMA's later examination of Microsoft's Inflection investment.
Competition analysis shifts beyond model quality toward the relationships linking model providers, large technology platforms and infrastructure suppliers, raising the salience of distribution and investment arrangements.

Third-order effects

If this approach persists, AI competition oversight will increasingly be organized around ecosystems and dependency relationships rather than only conventional product-market boundaries.
The eventual effect depends on how the CMA converts principles into cases: the subsequent inquiry into Google's Anthropic partnership suggests partnership structures may become a durable enforcement focus.

The trend: AI regulation is evolving from broad foundation-model principles toward competition oversight of the partnerships, distribution channels and infrastructure that shape access to AI markets.

Discussion

@stokel @stokel on x
This is also whiplash optics given the government has been going around the sector saying “we're light touch! the AI Singapore! come to us!”, while a week ago the same body now looking at AI blocked a $69 billion mega-merger https://www.gov.uk/...
@matthewlesh Matthew Lesh on x
OpenAI is a new company that literally had zero users just months ago. They're now competing with Google Search. Seems to be a quite well functioning competitive market.
@maxvonthun Max von Thun on x
Delighted to see the CMA moving quickly to understand the potential threats to competition from the AI race. Competition regulators were far too slow to stop Web 2.0 from being cornered by a few giants. With AI, we still have the chance to keep markets open and contestable. https…
@cmagovuk @cmagovuk on x
Our goal is to help this new, rapidly scaling technology develop in ways that ensure open, competitive markets and effective consumer protection. Read more: https://www.gov.uk/...
@daviddavismp David Davis on x
AI has huge potential to disrupt all areas of our lives - both for good and for bad. That is why this review is much needed. We have to find a way to keep it under control, ensuring we harness its potential but avoid the serious dangers posed. https://www.ft.com/...
@s8mb @s8mb on x
This seems... really premature. Fair enough to do a market study on AI, but the sector will change a lot between now and the end of the study in a year, and presumably well beyond that. How can the CMA draw any strong conclusions at this stage? https://www.ft.com/...
@s8mb @s8mb on x
Here's what they're doing - I was wrong to interpret this as a formal market study. Still, beyond learning about the tech (which is good) I question how much they can conclude about the structure of the market. https://www.gov.uk/...
@matthewlesh Matthew Lesh on x
The CMA is drunk with power. These antics are making the UK inhospitable for tech investment and hurting innovation. https://www.ft.com/...
@cmagovuk @cmagovuk on x
We've launched an initial review of artificial intelligence models: https://www.gov.uk/... https://twitter.com/...
@drphilipmarsden @drphilipmarsden on x
CMA AI Review will assess “the real opportunities” but also “what kind of guardrails, what principles, we should be developing in terms of ensuring that competition is working effectively [and] consumers are being protected”. https://twitter.com/...
@kevincollier@mastodon.social Kevin Collier on mastodon
Media is an extremely difficult business model. And I respect that Khan wants to speak through the paper of record. But since this is important writing from a current US official, here's a reminder that viewing links via archive.is gets you around most paywalls. …
@azeem Azeem Azhar on x
Sensible and measured messages from @linakhanFTC https://www.nytimes.com/...
@mat @mat on x
This interview with Geoffrey Hinton is truly 😱 It's “quite conceivable that humanity is just a passing phase in the evolution of intelligence” and “it may keep us around for a while to keep the power stations running, but after that, maybe not” https://www.technologyreview.com/ .…
@stevesi Steven Sinofsky on x
Lina Khan: We Must Regulate A.I. Here's How. https://www.nytimes.com/... // But how? Piece doesn't say. Also perhaps big tech should stop asking to be regulated (generally) though maybe suggest how so this back and forth can be concrete/cards on table.
@s8mb @s8mb on x
Market concentration is a particularly odd thing to worry about when the two most impressive consumer-facing AI services, ChatGPT and Midjourney, are from firms that have no incumbency prior to this. ChatGPT is the fastest-growing consumer app ever, with zero prior customer base.
@jathansadowski Jathan Sadowski on x
Many problems with tech stem from this still being a radical thing to say: “Although these tools are novel, they are not exempt from existing rules, and the FTC will vigorously enforce the laws we are charged with administering, even in this new market.” https://www.nytimes.com/.…
@mattinthemittel Matthew Mittelsteadt on x
A few initial thoughts on @linakhanFTC's recent @nytimes op-ed on AI regulation. In the article, she highlights several points of emphasis, below are my thoughts on each:🧵https://www.nytimes.com/ ...
@neil_chilson @neil_chilson on x
@elidourado “WHAT IF AI TURNS OUT TO BE AS BAD AS GETTING FREE, ENTERPRISE-QUALITY SERVICES IN EXCHANGE FOR TARGETED ADS?” https://twitter.com/...
@lehogg Luke Hogg on x
When potential algorithmic collusion by Big Bowling is the worst conceivable example, you know regulators are grasping at straws trying to justify sweeping regulation of AI https://www.nytimes.com/... https://twitter.com/...
@ylecun Yann LeCun on x
In case you are wondering: I have no doubt that superhuman AI systems will eventually exist. But today they don't, and we don't even have a basic design for them. So discussing how to make them safe is a bit like discussing how to make transatlantic flights safe in 1920.
@sarthakgh Sar Haribhakti on x
There's a comical worldview held by many in punditry and political circles that our “data” is like expensive watches or jewelry locked in safes and is now being stolen by companies like FB They couldn't care less about whether you personally are into Succession or poop memes http…
@oablanchard Olivier Blanchard on x
Background: Lina Khan (FTC Chairperson) is throwing everything she can at “Big Tech” under the guise of protecting competition and giving small companies a level playing field where they are able to innovate and compete with big tech companies. ... 🧵 https://twitter.com/...
@fabiochiusi Fabio Chiusi on x
(It doesn't really say “how”, though) https://www.nytimes.com/...
@datasociety @datasociety on x
“The trajectory of the Web 2.0 era was not inevitable,” @linakhanFTC writes. It was shaped by policy choices. “As the use of AI becomes more widespread, public officials have a responsibility to ensure this hard-learned history doesn't repeat itself.” https://www.nytimes.com/...
@accountabletech @accountabletech on x
Generative AI is turbocharging fraud, accelerating disinformation, and enabling discrimination. @LinaKhanFTC writes in the @nytimes how the @FTC is committed to upholding existing rules to rein in AI and promote fair and open competition. https://www.nytimes.com/...
@s8mb Sam Bowman on x
Many calls to regulate AI dwell on risks that we can deal with as they arise, like misinfo, or risks that are contradicted by experience, like market concentration, while ignoring the risks of weaponisation that we may not be able to fix after the fact. https://www.nytimes.com/..…
@jason_kint Jason Kint on x
“It may not be ready to replace professional writers, but it can already do a vastly better job of crafting a seemingly authentic message than your average con artist.” Which coupled with pervasive user data collection is good enough for ad targeting. https://www.nytimes.com/...
@linakhanftc Lina Khan on x
Although AI tools are novel, they are not exempt from existing rules. In @nytimes I explain how @FTC will pursue our mandate to promote fair competition and prevent unfair or deceptive practices in this new market. https://www.nytimes.com/...
@elidourado Eli Dourado on x
While Yudkowsky is warning us about AI killing everyone, Lina Khan suggests that the AI era could be as chaotic as having social media companies sell ads based on personal data. LMAO. https://www.nytimes.com/... https://twitter.com/...
@finmoorhouse Fin Moorhouse on x
New White House announcement on AI today. Some promising signs: interest in the overlap with biosecurity, improving infosec for top labs, and some kind of initiative where labs are committing to model evaluations. https://www.whitehouse.gov/... https://twitter.com/...
@sebkrier Séb Krier on x
White House announces an independent commitment from leading AI labs like Anthropic, Google, Hugging Face, Microsoft, NVIDIA, OpenAI, and Stability AI, to participate in a public evaluation of AI systems on an evaluation platform developed by Scale AI. https://www.whitehouse.gov/…
@rowancheung Rowan Cheung on x
Big announcements in the world of AI today from Deepmind, The US White House, Portfolio Pilot, and Crypto com. Here's the rundown on everything you need to know:
@pegobry Pascal-Emmanuel Gobry on x
The White House's safe AI: [Image of the Terminator with a speech bubble saying “As an AI language model, I cannot say something that would be offensive to certain groups"] https://twitter.com/...

Chronicles