There was a time when competition lawyers could explain monopoly power largely through familiar things.
Price.
Production.
Market share.
Distribution.
Exclusive contracts.
Control of physical infrastructure.
If one company owned the railway, the port, the pipeline or the only commercially viable route to market, the competitive problem was relatively easy to visualise.
Artificial intelligence is creating a rather stranger kind of infrastructure.
The railway may now be an operating system.
The port may be a search engine.
The pipeline may be a dataset.
And the tollgate may be the button, voice command or default setting through which a human being summons an artificial intelligence assistant.
That is why a seemingly technical decision made in Brussels on 16 July 2026 deserves considerably more attention than it has received.
The European Commission adopted two sets of binding specification measures against Google under the Digital Markets Act, Regulation (EU) 2022/1925. One set concerns interoperability between competing artificial-intelligence services and Google’s Android operating system. The other concerns access by competing search providers—including qualifying AI chatbots with search functionality—to Google Search data.
A week later, on 23 July 2026, the Commission imposed a combined €890 million in fines on Google in two separate Digital Markets Act non-compliance decisions: €460 million concerning self-preferencing within Google Search and €430 million concerning restrictions on the ability of app developers to steer users towards alternative purchasing channels outside Google Play.
The €890 million penalty was not an AI fine.
That distinction is important.
But the two developments belong to the same larger legal story.
Europe has moved beyond asking whether digital platforms charge too much.
It is increasingly asking a different question:
Does control of a digital gateway allow its owner to determine which competitors are technologically capable of reaching the customer at all?
Artificial intelligence makes that question urgent.
Because if AI assistants become a principal interface between humans and their phones, applications, search engines, shops, banks and digital services, whoever controls the operating system beneath the assistant may possess an extraordinary competitive advantage.
The next great antitrust disputes may therefore concern something much more fundamental than price.
They may concern who owns the doorway through which artificial intelligence enters everyday life.
The AI assistant is becoming an operating layer
For the moment, many people still experience generative AI as a destination.
They open an application.
They visit a website.
They type a prompt.
They wait for an answer.
That model is already changing.
The emerging AI assistant is less like an application and more like an intermediary between the user and everything else on the device.
You may ask it to:
“Tell my husband I’ll be twenty minutes late.”
“Book the taxi I normally use.”
“Find the invoice Zahid sent last Tuesday.”
“Translate what’s written on my screen.”
“Order my usual dinner.”
“Turn the brightness down.”
“Read my next appointment.”
“Reply to this email politely.”
“Find the cheapest flight to Dubai next week.”
“Compare these three insurance policies.”
At that point the assistant is not merely another application.
It has become an interface to the operating system, the user’s applications and increasingly the commercial world beyond them.
The Commission expressly recognised this dynamic in its July 2026 Android decision. It described AI assistants as becoming a primary way in which users interact with mobile devices and observed that Google’s Gemini benefits from deep integration with Android, while competing assistants historically lacked equivalent access to important operating-system features.
That is where competition law enters.
Competition increasingly depends upon technical capability
Suppose two AI assistants are equally intelligent.
One can be summoned from anywhere on the phone simply by saying a wake phrase.
The other requires the user to unlock the device, find an icon and open the application.
One can read what is currently displayed on the screen.
The other cannot.
One can interact with the user’s calendar, messaging app, transport application and device settings.
The other must send the user back to those applications manually.
One can perform actions across applications.
The other can merely tell the user how to perform them.
One has efficient access to on-device processing resources.
The other operates under greater technical constraints.
Which assistant will the average consumer perceive as “better”?
Almost inevitably, the first.
But suppose the difference does not arise because the first AI model is intrinsically superior.
Suppose it arises because the company supplying it also controls the operating system.
That is no longer merely a product-design question.
It becomes a competition question.
The distinction is subtle but essential:
competition law should not normally punish a company for building a better product.
It may, however, become concerned where a vertically integrated platform uses control of an indispensable or strategically important layer to deny rivals equivalent opportunities to compete.
Article 6(7) of the Digital Markets Act attacks precisely that bottleneck
Article 6(7) of the DMA requires designated gatekeepers to provide providers of services and hardware, free of charge, with effective interoperability and access for interoperability purposes to the same operating-system, hardware and software features available to the gatekeeper’s own services, subject to strictly necessary and proportionate measures protecting system integrity.
The underlying reasoning is straightforward.
A company controlling an operating system may simultaneously operate services which compete with third-party services dependent upon that operating system.
That places the company in a dual position.
It is both:
the owner of the playing field; and
a player standing on it.
The Digital Markets Act expressly recognises the competitive danger created by such vertically integrated gatekeeper positions. Its recitals explain that withholding equivalent access to operating-system, hardware or software functions can undermine innovation and consumer choice and therefore justify interoperability requirements.
Artificial intelligence makes that previously abstract principle remarkably concrete.
What Europe is requiring Google to open
The July 2026 decision concerns 11 Android features regarded by the Commission as important to effective AI competition.
They fall broadly into four categories: invocation; contextual understanding; actions across applications and the operating system; and access to hardware and software resources.
Take invocation first.
Competing AI assistants should be able to make use of important system-wide access points rather than having strategically valuable routes reserved to Google’s own services.
The Commission’s measures contemplate third-party assistants being capable of invocation using customised wake words and through centrally located access points such as the long-press home-button or navigation-handle function.
In simple language:
“Hey Google” should not become the legal equivalent of owning the front door.
If voice interaction becomes one of the principal ways humans summon AI, then exclusive access to always-on voice invocation can itself become competitively significant.
Context may be even more valuable than invocation
A useful AI assistant needs context.
It needs to know what the user is doing.
What is on the screen?
Which app is open?
What was said in the conversation?
Where is the user?
What information is stored in relevant applications?
Without appropriate access, the assistant is substantially less capable.
The Commission’s measures therefore address access by competing AI services to contextual information, application data stored on the device, relevant screen content and other technical inputs necessary to deliver sophisticated assistance.
That creates a striking competition-law proposition.
Access to context may itself become an input necessary for effective competition.
The AI model may be brilliant.
But if it cannot see what the user sees, understand the device’s context or interact with the environment in which it operates, intelligence alone may not make it competitive.
The assistant also needs hands
The next category concerns action.
A chatbot that can answer a question is useful.
An agent which can actually do something may be dramatically more valuable.
The July measures therefore address the capacity of competing AI services to interact with applications and perform tasks on the user’s behalf.
The Commission gives examples including drafting and sending emails, adding items to shopping lists, ordering food, interacting with ride-sharing services and adjusting operating-system settings.
This is where the distinction between an AI chatbot and an AI agent becomes legally consequential.
A chatbot speaks.
An agent acts.
Once software can act across other applications, control of the permissions and interfaces permitting those actions becomes a competitive resource.
The operating-system owner potentially decides which digital agent is powerful and which is merely conversational.
Competition law therefore moves downwards into architecture.
The battlefield is becoming infrastructure
That is the wider point.
Traditional competition disputes often ask whether competitors have access to customers.
AI competition increasingly asks whether competitors have access to capabilities.
Can the AI hear?
Can it see the user’s context?
Can it invoke itself?
Can it act?
Can it use the necessary processing resources?
Can it communicate with other applications?
Can it remain responsive?
Can it obtain the data required to improve?
Those questions sound like engineering.
They are rapidly becoming law.
Google Search data creates the second battlefield
The Commission’s second July 2026 specification decision concerns something less visible but potentially just as important:
search data.
Article 6(11) of the DMA requires designated gatekeeper search engines to provide competing search undertakings, upon request, access on fair, reasonable and non-discriminatory terms to certain ranking, query, click and view data generated by users, with personal data being anonymised.
Why does that matter?
Because a search engine improves partly by observing enormous volumes of human behaviour.
What did people search?
Which result did they click?
Which result did they ignore?
Did they reformulate the query?
Which ranking performed well?
Where did users go afterwards?
Aggregated across enormous volumes of searches, this behavioural information can improve relevance and ranking.
The DMA itself recognises that the scale at which a dominant search engine accumulates ranking, query, click and view data can create a significant barrier to entry and expansion for competitors.
That produces another uncomfortable competition question.
If the best way to build a competitive search engine is to possess search data, and the best way to accumulate search data is already to operate a successful search engine at enormous scale, how does a new rival ever catch up?
This is the circularity of data advantage.
Success generates data.
Data improves the product.
The improved product attracts users.
Users generate more data.
And the cycle repeats.
AI chatbots have now entered that search-data argument
This is where the July 2026 decision becomes especially significant for artificial intelligence.
The Commission expressly states that AI chatbots offering search functionality may qualify to receive shared Google Search data under the specified framework.
That is a major conceptual development.
The old legal distinction between “search engine” and “AI assistant” is becoming unstable.
If a user asks an AI:
“Which restaurants near my hotel serve kosher food after 10 pm?”
has the user conducted a search?
If the AI retrieves current information from the web, ranks candidate answers and gives one recommendation, the economic function begins to resemble search even though there is no traditional results page.
Competition regulation therefore has to follow functionality rather than product labels.
The relevant question becomes:
what service is the AI actually performing?
Search may disappear behind the answer
This transition has enormous commercial consequences.
Traditional internet search displays alternatives.
The user sees perhaps ten blue links, advertisements, maps, snippets and other results.
An AI assistant may instead provide:
“Book Restaurant X. It is the best option.”
That changes the economics of ranking.
When twenty results are displayed, being ranked fourth may still be commercially valuable.
When the AI communicates only one answer, the distinction between first and second can become existential.
The DMA itself anticipated this broader issue. Its recitals explain that ranking includes forms of relative prominence such as display, linking and even voice results, and expressly contemplates circumstances in which a platform communicates only one result to the end user.
That language now appears remarkably prescient.
The future gatekeeper may not merely control the list.
It may control the answer.
AI therefore makes self-preferencing more dangerous
Self-preferencing sounds technical until we imagine it through AI.
Suppose a user tells an assistant:
“Find me a hotel.”
The assistant’s owner operates a hotel-booking service.
Which booking service does the assistant favour?
Or:
“Buy me concert tickets.”
The owner also operates a marketplace.
Or:
“Find a lawyer.”
The platform sells advertising or directory placement to lawyers.
Or:
“Order groceries.”
The platform owns grocery services.
Or:
“Book the cheapest flight.”
The platform offers travel products.
When AI increasingly mediates transactions, there may be no visible rankings for the user to inspect.
The possibility of discriminatory treatment becomes less obvious precisely because the interface becomes more convenient.
The user may not even know which alternatives were excluded.
This is where the €890 million fine becomes relevant
Again, the July 23 fine did not concern AI interoperability.
The Commission found two separate DMA infringements.
First, it concluded that Google gave preferential treatment in Search to its own services, including shopping, hotels, transport and sports results, including through more prominent presentation and enhanced visual treatment unavailable to comparable third-party services.
For that conduct, the Commission imposed a €460 million fine.
Second, the Commission concluded that Google had not complied with the DMA’s anti-steering obligations applicable to Google Play because developers were restricted in their ability to communicate alternative offers and conclude transactions with users through channels of their choice.
That infringement attracted a further €430 million fine.
The combined penalty was therefore €890 million.
The significance for AI lies not in pretending those were AI infringements.
It lies in observing the regulatory philosophy.
Europe is plainly willing to intervene where a gatekeeper uses control of one layer of a digital ecosystem in ways the Commission considers to disadvantage businesses operating through that layer.
The Android AI interoperability decision extends the same structural concern directly into the emerging AI economy.
The fine also tells AI companies that the DMA has teeth
The DMA is not a voluntary code of digital etiquette.
Article 30 permits the Commission to impose fines of up to 10 per cent of a gatekeeper’s total worldwide turnover in the preceding financial year for specified infringements, including failures to comply with obligations under Articles 5, 6 and 7 or binding specification measures.
For certain repeated infringements, the ceiling may rise to 20 per cent of worldwide turnover.
That matters for AI interoperability because the July specification measures were adopted pursuant to the DMA’s legally binding compliance machinery.
The dispute is therefore no longer:
“Would it be nice if rival AI assistants had better Android access?”
It is:
“What access does European law require a designated gatekeeper to provide?”
That is an entirely different conversation.
Europe is regulating architecture before proving individual harm in every case
The Digital Markets Act also represents a departure from conventional competition enforcement.
Traditional antitrust litigation can be painstakingly slow.
A regulator may have to define the relevant market, establish dominance, prove abusive conduct, establish anti-competitive effects, litigate economic evidence and survive appeals.
Digital markets may move faster than the case.
By the time the court establishes that a particular digital practice foreclosed competitors, the competitor may no longer exist.
The DMA attempts to address that problem through ex ante obligations imposed upon designated gatekeepers.
In other words, instead of repeatedly asking whether a particular gatekeeper practice constitutes an abuse under general competition law, the legislation establishes behavioural duties in advance for firms falling within its gatekeeper architecture.
That is why the July 2026 AI proceedings are so significant.
Europe is not waiting for a future AI rival to collapse and then conducting a ten-year historical inquiry into whether inaccessible Android functions contributed to the collapse.
It is attempting to specify interoperability requirements while the AI-assistant market is still developing.
But forced interoperability is not an easy principle
There is an obvious counterargument.
If a company creates an operating system, develops sophisticated integrated features and invests billions in hardware, software and artificial intelligence, why should competitors automatically be entitled to use those capabilities?
That concern should not be dismissed.
Competition policy which becomes too aggressive can punish innovation.
If the law tells every successful technology company that innovations it develops must immediately be supplied to rivals, it may weaken incentives to invest.
There is also a fairness intuition here.
A company which builds something valuable may reasonably say:
“My competitor is entitled to compete with me. It is not necessarily entitled to compete using everything I built.”
That proposition has considerable force.
Competition law must therefore distinguish between protecting competition and engineering competitors.
They are not the same thing.
The difficulty lies in the gatekeeper’s dual role
The answer Europe gives is essentially structural.
The intervention is not based simply upon the proposition that Google’s Android technology is valuable.
The concern arises because Android is a designated core platform service through which other businesses reach end users while Google simultaneously provides services competing with those businesses.
The Commission says approximately 60 per cent of EU mobile users use Android devices, making Android an especially important distribution channel for AI services.
The regulatory theory is therefore not:
“Gemini is successful, so competitors deserve help.”
It is closer to:
“Where the owner of an exceptionally important gateway competes against businesses dependent upon that gateway, it should not reserve indispensable gateway capabilities to itself.”
Whether one agrees with the policy or not, that is a much more defensible statement of the issue.
Security is the strongest legitimate objection
Opening deep operating-system functions to third-party AI services creates obvious risks.
Consider what the Commission wants competing assistants to be capable of doing.
Listening for wake words.
Reading screen context.
Interacting with applications.
Accessing device resources.
Performing tasks.
Potentially examining data necessary to understand user intent.
These are powerful permissions.
A poorly designed assistant could create security and privacy risks.
A malicious assistant could be considerably worse.
The DMA therefore does not require interoperability without qualification. Article 6(7) permits strictly necessary and proportionate measures protecting operating-system, virtual-assistant, hardware and software integrity where properly justified.
The Commission likewise emphasises that its July measures contain safeguards directed at user privacy, device integrity and security.
This will be one of the most important lines of future litigation and regulatory argument.
When is a security restriction genuine?
And when is “security” merely a convenient vocabulary for keeping competitors outside?
Every interoperability dispute can become a security dispute
This is not a cynical observation.
Platform security is real.
But incentives matter.
The company controlling a platform has legitimate reasons to protect users.
It may simultaneously have commercial reasons to restrict rival access.
The regulator therefore confronts a difficult evidential problem.
Suppose a platform says:
“Third-party assistants cannot access this feature because doing so creates cybersecurity risk.”
A competitor replies:
“Your own assistant accesses precisely the same feature.”
The platform responds:
“We can verify our own code and therefore manage the risk differently.”
That may be technically correct.
Or strategically convenient.
Or both.
Competition regulation must distinguish necessity from foreclosure.
That task is difficult precisely because the most important evidence may be deeply technical.
Privacy creates an equally complicated conflict
Search-data sharing presents the mirror-image problem.
Competition policy says:
Share valuable data so competitors can challenge the incumbent.
Privacy law says:
Do not disclose identifiable information about users without lawful justification and safeguards.
The July Google Search specification decision attempts to reconcile these principles through anonymisation requirements.
The Commission states that Google should share the same relevant search data it collects for optimisation of its own search service, subject to anonymisation, and has designed a multi-layered anonymisation methodology informed by privacy considerations. It also permits Google to assess serious cybersecurity or data-protection risks posed by sharing with a particular recipient.
This is an extraordinary example of one regulatory objective colliding with another.
Competition wants data to move.
Privacy sometimes wants data not to move.
Modern technology law increasingly consists of deciding which principle prevails, to what extent, and under which safeguards.
Data is not merely the new oil
The cliché that “data is the new oil” is now exhausted.
Data behaves differently from oil.
Oil is consumed when used.
Data can be copied.
Oil does not ordinarily become more valuable because millions of other people have consumed the same barrel.
A behavioural dataset may become more useful precisely because of scale.
Data can also possess value which only becomes apparent when combined with other datasets or computational capability.
The AI economy therefore complicates traditional scarcity.
A dataset can be simultaneously:
commercial property,
an input into competition,
a privacy risk,
training material,
an evidential record,
a cybersecurity asset,
and a source of market power.
Competition law cannot deal with data as though it were simply another warehouse of commodities.
The truly scarce asset may be feedback
AI competition also requires us to distinguish data from feedback loops.
A model can theoretically be copied.
Software can be licensed.
But an incumbent receiving enormous volumes of real-world user interaction may continuously learn what users actually want.
Which answers satisfy them?
Which results do they reject?
Which query reformulations indicate failure?
Which actions do they complete?
Which recommendation leads to a purchase?
That feedback can improve the product.
The advantage therefore comes not simply from possessing historical information.
It comes from occupying the position through which future information continuously arrives.
That is why distribution matters so profoundly to AI competition.
Defaults may become more powerful than price
Digital markets have taught us repeatedly that defaults matter.
Most users do not investigate every available setting.
They use what is installed.
They click what is visible.
They accept what works.
AI may intensify that behaviour.
If the default assistant can perform almost every useful task, how many consumers will install, configure and grant extensive permissions to a competitor?
This produces another competition-law puzzle.
Suppose switching is technically possible.
Is that enough?
Or must switching also be realistically usable?
The DMA expressly attempts to prevent artificial technical barriers which make switching ineffective and requires gatekeepers to respect end-user choice between competing services.
The legal meaning of choice may therefore expand.
Nominal choice is not necessarily effective choice.
The wake word may become the twenty-first century supermarket shelf
Competition law once worried about access to retail shelf space.
If one company controlled the shelf, rival products might never reach consumers.
AI creates an equivalent problem in invisible form.
The valuable shelf may be:
the default assistant,
the voice invocation privilege,
the home-screen gesture,
the operating-system permission,
the ranking algorithm,
the AI recommendation,
or the system API.
These things do not look like marketplaces.
But economically they can determine whether a marketplace exists.
That is why lawyers advising technology companies must become comfortable discussing product architecture.
Technical design can embody market power.
The most valuable business may be the one the user never consciously opens
There is another reason the Android decision deserves attention.
The old application economy rewarded visibility.
Businesses wanted consumers to open their application.
AI agents may reverse that.
Suppose a consumer tells the assistant:
“Order the same groceries as last week.”
The assistant interfaces with a supermarket service in the background.
The user may never open the supermarket application.
Or:
“Send the money to Ali.”
The AI invokes a payment application.
Or:
“Book me a car.”
The AI invokes a ride-hailing platform.
The AI becomes the customer’s immediate interface.
The underlying application becomes infrastructure.
This creates a new competitive hierarchy.
The company controlling the agent layer may acquire enormous influence over businesses beneath it.
AI agents could become universal intermediaries
That is perhaps the largest long-term implication.
Search engines mediate information.
App stores mediate software distribution.
Marketplaces mediate sellers and buyers.
Payment systems mediate transactions.
Social networks mediate communication.
An advanced AI agent could mediate all of them.
One interface might decide which search result to trust, which app to invoke, which shop to use, which payment method to select and which advertisement—or commercial recommendation—to present.
At that stage, competition authorities will confront an extraordinary concentration of intermediary power.
The agent may not merely help the consumer make decisions.
It may make decisions for the consumer.
The jurisprudence of recommendation is therefore likely to become one of the great competition-law subjects of the next decade.
When does convenience become gatekeeping?
Users may genuinely prefer an integrated ecosystem.
That preference must matter.
A regulatory system should not deliberately make technology worse simply because integration produces market power.
If Gemini works especially well because Google designed Android and Gemini together, forcing artificial separation could harm consumers.
The goal should not be technological fragmentation for its own sake.
But integration should also not become an automatic defence.
The legal enquiry should ask:
Did consumers choose the integrated service because it was superior?
Or did the service appear superior because competitors were prevented from obtaining access necessary to provide equivalent functionality?
Those two realities can look identical from the user’s perspective.
Competition law exists partly to determine which one occurred.
There is a danger of regulators designing products
Europe should nevertheless proceed carefully.
Detailed interoperability regulation risks placing public authorities in the position of technology architects.
The July decision addresses matters as specific as wake-word invocation, contextual inputs, application actions and access to on-device resources.
That level of specificity may be unavoidable.
But it carries risks.
Regulators do not innovate at the same speed as technology companies.
A mandated technical solution could become obsolete.
Interoperability requirements may create unforeseen vulnerabilities.
Compliance engineering may consume resources otherwise used for innovation.
Competitors may lobby for access rather than develop alternatives.
The administrative process may eventually become a battlefield for strategic advantage.
Good competition regulation therefore requires humility.
The State should open bottlenecks where justified without believing it can design the digital economy better than those building it.
The opposite risk is considerably worse: regulating after the market has tipped
There is, however, a reason Europe is willing to intervene early.
Digital markets can “tip”.
Network effects, data accumulation, installed-device bases, defaults and ecosystem integration can make early advantages self-reinforcing.
Once a particular AI assistant becomes the habitual intermediary for hundreds of millions of users, switching may become increasingly difficult.
Businesses integrate around it.
Developers optimise for it.
Users grant it permissions.
The assistant learns preferences.
Historical data accumulates.
Commercial partnerships develop.
The cost of leaving rises.
By the time a conventional competition case concludes that exclusion occurred, the emerging market may already have hardened.
The policy dilemma is therefore genuine:
intervene early and risk regulating a market still evolving;
or
intervene late and discover there is no longer a competitive market to preserve.
There is no intellectually painless answer.
This matters to AI start-ups everywhere
A small AI company may imagine its competitive challenge is building a sufficiently capable model.
That is only part of the problem.
It must also ask:
Can users find us?
Can users invoke us conveniently?
Can we obtain the permissions necessary to perform useful tasks?
Can we integrate with applications?
Can we access suitable computing resources?
Can we acquire the data needed to improve?
Can users make us their practical default?
Can we reach customers without paying a vertically integrated competitor for access?
Can we appear in recommendations made by another AI?
These are questions about market structure.
A technically brilliant model may fail commercially because it cannot obtain access to the surrounding ecosystem.
The new competition lawyer therefore needs to understand the distinction between model capability and ecosystem capability.
The same issue will emerge in enterprise AI
The consumer smartphone is merely one example.
Enterprise software may face similar questions.
Imagine a dominant productivity suite containing its own AI assistant.
Third-party AI tools require access to documents, calendars, messages, spreadsheets and workflow systems to compete.
How much interoperability should be required?
What if a cloud provider favours its own models?
What if access to proprietary enterprise data is technically possible but priced prohibitively?
What if the platform’s own assistant receives privileged API access?
What if third-party assistants receive data with greater latency?
What if the relevant functionality is technically available but buried behind cumbersome user permissions?
Competition can be impaired by friction, not merely prohibition.
This is likely to become increasingly important.
An API can be legally “open” yet commercially useless
That point deserves emphasis.
A platform may say:
“We provide an API. Therefore competitors have access.”
But effective interoperability requires more than the theoretical existence of an interface.
Is the API documented?
Is it stable?
Does it provide equivalent functionality?
Is access timely?
Are usage limits reasonable?
Does the incumbent’s own service enjoy functionality unavailable through the API?
Are approval procedures discriminatory?
Are competitors subjected to additional user prompts which the incumbent avoids?
Does latency materially degrade performance?
Can the API be withdrawn?
Technical equality on paper can conceal competitive inequality in practice.
The Commission’s July measures therefore require interoperability to be effective and contemplate documented frameworks, technical assistance and equal access without unnecessary user friction.
Competition law is becoming a law of interfaces
This may ultimately be the most important jurisprudential development.
Industrial competition law often regulated relationships between firms.
Digital competition law increasingly regulates relationships between systems.
API to API.
Assistant to operating system.
Search engine to dataset.
Application to app store.
Advertiser to platform.
Device to service.
The interface between technologies becomes the place where market power is exercised.
A refusal, restriction or discriminatory condition imposed at that interface can determine whether a competitor has a viable business.
Law therefore moves into code.
What does all this mean for Pakistan?
Quite a lot.
Pakistan does not presently possess a direct equivalent of the EU Digital Markets Act’s gatekeeper regime.
But Pakistani competition law already contains principles capable of addressing exclusionary conduct.
Section 3 of the Competition Act 2010 prohibits abuse of a dominant position through practices which prevent, restrict, reduce or distort competition in a relevant market. The Competition Commission of Pakistan explains that exclusionary abuses can include refusal to deal or supply and conduct preventing market entry, while discriminatory practices can include imposing dissimilar conditions on equivalent transactions in a manner placing other parties at a competitive disadvantage.
Pakistan’s statutory framework also contains a rebuttable-looking legal architecture around dominance: the CCP explains that a market share exceeding 40 per cent gives rise to a statutory presumption of dominance, while an undertaking below that threshold may nevertheless be dominant where it can behave to an appreciable extent independently of competitors, customers, consumers and suppliers.
Importantly, dominance itself is not prohibited.
Abuse is.
That distinction is foundational.
A company should not be punished for becoming successful through innovation, efficiency or consumer preference.
The law intervenes where market power is used in prohibited ways.
Pakistan should watch the European experiment before copying it
There would be little wisdom in simply reproducing the Digital Markets Act in Islamabad.
The EU possesses a market of a vastly different scale, regulatory infrastructure developed over decades and a specific history of competition enforcement against global digital platforms.
Pakistan’s circumstances differ.
But the European experience should be studied carefully because the same economic structures can appear here.
Pakistani consumers increasingly transact through digital platforms.
Local businesses depend upon search visibility, mobile operating systems, social networks, marketplaces, payment networks, telecommunications services and application ecosystems controlled by larger enterprises.
Pakistan’s technology companies likewise depend heavily upon foreign platform infrastructure.
A Pakistani business may therefore experience digital gatekeeping even where the gatekeeper itself is not Pakistani.
Pakistan’s existing Competition Act may be more flexible than it appears
One strength of section 3 is that the list of abusive practices is not exhaustive.
The CCP describes abuse broadly as conduct by a dominant undertaking which prevents, restricts, reduces or distorts competition and recognises exploitative, exclusionary and discriminatory categories.
That conceptual breadth could potentially permit consideration of novel digital conduct without waiting for Parliament to name every future technological practice.
The difficult questions would be evidential and economic.
What is the relevant market?
Is it smartphone operating systems?
Mobile AI assistants?
Search-enabled AI assistants?
App distribution?
Search advertising?
Digital payments?
Cloud AI infrastructure?
Does a foreign platform’s conduct sufficiently affect competition in Pakistan?
What constitutes exclusion?
When is refusal of interoperability objectively justified?
What access remedy would be proportionate?
Those questions cannot be answered by mechanically importing European findings.
But Pakistani competition law has a vocabulary with which to begin asking them.
CCP itself is becoming more technologically sophisticated
The Competition Commission of Pakistan has meanwhile been modernising its enforcement tools.
In March 2026 the Commission publicised an Automated Digital Market Intelligence System designed to monitor online advertisements, social-media content and digital platforms, generate alerts and assist in identifying potentially misleading commercial practices.
The CCP has also stated in 2026 that it is developing guidance in emerging areas including e-commerce as part of its broader enforcement-modernisation efforts.
These developments currently focus more heavily upon monitoring and deceptive practices than upon an EU-style gatekeeper regime.
Nevertheless, they demonstrate an institutional recognition that competition enforcement increasingly requires regulators to understand digital markets in real time.
The next stage should include deeper analytical capacity concerning algorithms, platform dependence, digital ecosystems, network effects, interoperability, data advantage and AI-assisted commercial intermediation.
Pakistan may have more to lose from foreign gatekeeping than domestic gatekeeping
This is an important strategic point.
Many Pakistani policy discussions naturally focus upon how Pakistani companies should be regulated.
But digital economic power frequently sits elsewhere.
A Pakistani software developer may depend upon:
a foreign operating system;
a foreign app store;
a foreign cloud service;
a foreign search engine;
a foreign foundation model;
a foreign advertising network;
and a foreign social-media platform.
The commercial viability of the Pakistani business may therefore be influenced by rules written in Brussels, California or elsewhere because those jurisdictions regulate the platforms through which Pakistani companies reach international customers.
This makes international competition law a market-access issue for Pakistan.
A Pakistani AI company may benefit from European regulation without being European
Imagine a Pakistani technology company develops an excellent AI assistant capable of serving multilingual European customers.
Its ability to compete on Android devices improves because EU law requires greater interoperability.
The Pakistani company may indirectly benefit from European competition regulation.
Likewise, if an eligible search-enabled AI business can lawfully access relevant anonymised search data under FRAND terms, that regulatory intervention could affect businesses incorporated well beyond the European Union, depending upon the applicable eligibility and operational requirements. The Commission’s July decision expressly treats eligible AI chatbots with search functionality as potentially within the search-data-sharing framework.
This illustrates a recurring feature of European digital law.
The regulation may be territorial.
Its commercial effects are international.
Pakistani companies should therefore read platform regulation as business law
A common mistake is to treat the Digital Markets Act as something relevant only to European competition specialists.
For an international software company, it may determine:
which features are accessible;
what data can be obtained;
how an application can communicate with users;
whether alternative payment options may be promoted;
whether platform services can favour their own offerings;
and whether AI functionality can integrate with an operating system.
Those are commercial questions.
International businesses should therefore map foreign digital regulation alongside taxation, privacy, export controls, sanctions and intellectual-property law when planning cross-border products.
The same contractual consequences will arrive in Pakistan
European interoperability rules may also filter into contracts.
An AI developer may seek warranties from device or platform partners concerning API access.
Customers may require assurances concerning portability.
Agreements may address regulatory changes to platform functionality.
Technology licences may distinguish between access promised contractually and access mandated by law.
Data-sharing arrangements may need to address anonymisation, confidentiality, cybersecurity and restrictions upon re-identification.
Indemnities may allocate responsibility if data supplied under an interoperability regime creates downstream privacy exposure.
Companies should therefore avoid treating competition law solely as litigation.
Competition regulation increasingly changes contract design.
Acquisition due diligence must examine dependence upon gatekeepers
Suppose an investor is buying an AI company.
Revenue looks excellent.
The technology is impressive.
But 90 per cent of the product’s functionality depends upon privileged access to one platform API.
That API can be changed unilaterally.
The company has no contractual guarantee of continued access.
A competing service owned by the platform receives deeper integration.
That is not merely a technical risk.
It is a competition and valuation risk.
Due diligence should increasingly ask:
Which platforms are indispensable?
Which permissions are discretionary?
Which interfaces are regulated?
Which contracts govern access?
Could regulatory intervention improve or worsen the company’s position?
Does the target rely upon access which could disappear?
Does it possess data which regulators may require to be shared?
Is its business model vulnerable to platform self-preferencing?
Market power increasingly sits inside technical dependencies.
The future merger lawyer will need a systems diagram
AI transactions will therefore require lawyers to understand entire stacks.
Foundation model.
Cloud infrastructure.
Operating system.
Hardware.
Application.
Data.
Distribution.
Search.
Advertising.
Payment.
Agent layer.
Customer interface.
A conventional market-share analysis may not reveal where control actually resides.
A company with modest revenue in one market may control an infrastructure layer indispensable to several adjacent markets.
Competition analysis must therefore become more architectural.
Europe’s great gamble: contestability
The DMA repeatedly uses a concept less familiar to the ordinary consumer than “competition”:
contestability.
A market may contain several firms yet still be difficult to contest.
A new entrant theoretically can enter but practically cannot obtain the data, interfaces, customers or interoperability necessary to become meaningful.
The policy objective is therefore not merely maintaining multiple company names.
It is preserving the possibility that rivals can genuinely challenge incumbents.
That distinction is extremely important in AI.
We should not merely ask:
“How many AI assistants exist?”
We should ask:
“How many AI assistants can realistically compete for the user’s primary digital relationship?”
Those are different questions.
Five assistants are not competition if only one has the keys
Imagine five AI applications exist.
Four are more intelligent than the default.
But only the default assistant can:
listen continuously for a wake word;
control system settings;
access context;
perform actions across applications;
use on-device models efficiently;
and appear through privileged system-level invocation.
Counting five competitors would tell us very little.
The meaningful competition question is whether the competitors possess the capability to challenge the incumbent.
That is why interoperability has become central.
There is an ethical dimension to gatekeeping as well
Competition law is often described as economically technical.
At its best, however, it protects a form of commercial liberty.
It seeks to ensure that success is earned by serving consumers rather than by closing the doors through which rivals would otherwise compete.
That does not mean every business deserves success.
It means the contest should remain meaningful.
There is an old idea of fairness embedded here.
A powerful undertaking may legitimately build a magnificent marketplace.
The harder question arises when it also sells in the marketplace, determines who may enter, controls where competitors appear, owns the payment system and eventually appoints an AI assistant which decides what the customer should buy.
At some point, ownership of infrastructure becomes governance of other people’s commercial lives.
Law cannot ignore that transformation.
But neither should competition law become an entitlement to another company’s success
Fairness cuts both ways.
There is nothing just about forcing an innovator to subsidise a rival merely because the rival failed to build equivalent technology.
Nor should regulators turn interoperability into a general right to appropriate intellectual investment.
The legal challenge is therefore boundary-setting.
Which features are truly bottlenecks?
Which access restrictions are exclusionary?
Which are legitimate security measures?
Which data genuinely must be shared for contestability?
What compensation is appropriate?
What information can safely be anonymised?
When does interoperability undermine innovation?
When does refusing interoperability undermine competition?
Reasonable people can disagree about those lines.
That disagreement is precisely why the July 2026 decisions matter.
Europe is beginning to draw them.
The €890 million question is ultimately larger than Google
It is easy to turn every Big Tech story into a morality play.
Google bad.
Regulator good.
Or the opposite:
Innovator good.
Bureaucrat bad.
Neither frame is intellectually useful.
Google has created enormously valuable products used voluntarily by billions of people.
Its success cannot simply be presumed unlawful.
Europe nevertheless confronts a legitimate structural concern: companies controlling essential digital gateways can acquire the ability to privilege adjacent businesses which they also own.
The DMA is an attempt—controversial, ambitious and still experimental—to address that structural problem before it becomes irreversible.
Artificial intelligence dramatically raises the stakes.
The next monopoly may not charge the consumer anything
This is why price will increasingly become an inadequate measure of market power.
Search is often free to the user.
AI assistants may be bundled into devices or services.
Operating systems may generate value elsewhere in the ecosystem.
A consumer may pay zero and still be affected by reduced competition.
The harm may instead appear as:
less innovation;
reduced privacy;
poorer quality;
fewer choices;
higher prices charged to business users;
preferential treatment of affiliated services;
reduced opportunity for competitors;
greater dependence upon one ecosystem;
or control over what information and commercial choices the user sees.
Competition law has therefore had to become less obsessed with the retail price appearing on the invoice.
The most powerful gatekeeper may be invisible to the user
This may be especially true of AI.
A brilliant intermediary becomes invisible precisely because it is convenient.
The user asks.
The assistant decides.
The transaction happens.
Behind that apparent simplicity may be an enormous chain of commercial choices.
Which search provider supplied the information?
Which retailer was considered?
Which advertisements influenced the recommendation?
Which payment provider was selected?
Which app was invoked?
Which businesses were omitted?
Why?
When users stop consciously navigating digital markets, the entity navigating on their behalf acquires immense power.
That power will attract competition scrutiny.
Regulators will eventually have to examine AI recommendations themselves
Today’s interoperability decision concerns whether rival AI systems can compete.
Tomorrow’s disputes may concern how those systems compete inside their own answers.
Does an assistant favour affiliated products?
Does payment influence recommendations?
Are sponsored answers identifiable?
Does the assistant systematically exclude businesses refusing certain commercial terms?
Are results ranked objectively?
Does the AI steer transactions towards platforms owned by the same corporate group?
Can a business challenge its disappearance from AI recommendations?
Can regulators audit the underlying algorithm?
These questions are coming.
The old search-results page merely made them easier to see.
“Default AI” could become more valuable than “default search”
For twenty years, competition authorities argued over default search engines.
The next argument may concern the default intelligence layer.
A search engine answers search queries.
An AI assistant may eventually:
search,
communicate,
shop,
navigate,
bank,
schedule,
draft,
translate,
recommend,
and transact.
If that occurs, the commercial value of becoming the user’s default AI intermediary could dwarf the value of a default search setting.
Competition authorities understand this possibility.
The July 2026 Android intervention should be read in that light.
Europe is attempting to preserve assistant-level competition before the role of the assistant becomes fully entrenched.
What international businesses should do now
Businesses operating within digital ecosystems should identify how dependent their products are upon gatekeeper infrastructure.
Technology companies should understand which operating-system functions, APIs, datasets, app-store rules and search services their products require.
AI developers should distinguish between model quality and ecosystem accessibility.
Companies entering European markets should determine whether DMA rights may improve their ability to interoperate with designated platforms.
Businesses providing search-enabled AI services should carefully examine whether the July 2026 Google Search data-sharing framework could become relevant to them.
Platform operators should review whether vertically integrated structures create self-preferencing or discriminatory-access risks.
Investors should examine technical dependency during due diligence.
And boards should treat interoperability and platform-access risks as commercial governance issues rather than leaving them exclusively to engineering teams.
Frequently Asked Questions
Did the EU fine Google €890 million because of artificial intelligence?
No. The 23 July 2026 fines concerned two separate DMA infringements: self-preferencing of Google’s own services in Google Search and restrictions concerning developers steering consumers towards alternative purchasing channels outside Google Play. The Commission imposed €460 million and €430 million respectively.
Why is the €890 million fine relevant to AI competition?
Because it demonstrates the Commission’s willingness to use the same Digital Markets Act enforcement framework against gatekeeper behaviour. Separately, on 16 July 2026 the Commission adopted binding measures requiring Google to provide interoperability between Android and competing AI services and to improve access to Google Search data for eligible rival search providers.
What does Article 6(7) of the DMA require?
In broad terms, designated gatekeepers must provide providers of services and hardware with effective, free interoperability with and access to relevant operating-system, hardware and software features available to the gatekeeper’s own services, subject to permitted proportionate measures protecting system integrity.
What Android functionality does the July 2026 decision concern?
The measures concern 11 Android features grouped around invocation of AI services, access to user/device context, ability to perform actions in applications and the operating system, and access to relevant hardware and software resources.
Can another AI assistant use its own wake word on Android?
The Commission’s July measures envisage third-party AI assistants being capable of voice invocation using their own wake words and accessing other system-level invocation mechanisms, subject to implementation and security safeguards.
Does Google have to share Search data with competitors?
Article 6(11) requires access on fair, reasonable and non-discriminatory terms to specified ranking, query, click and view data for qualifying competing search providers, with personal data anonymised.
Can an AI chatbot qualify for Google Search data?
The Commission’s July 2026 specification decision expressly states that AI chatbots offering search functionalities may qualify for the data-sharing framework.
Does that mean Google must disclose users’ personal search histories?
No. The DMA requires relevant personal data to be anonymised, and the July decision contains specific anonymisation and cybersecurity/data-protection safeguards.
How large can DMA fines become?
For relevant infringements the DMA permits fines of up to 10 per cent of a gatekeeper’s worldwide annual turnover, with the ceiling potentially rising to 20 per cent for specified repeated infringements.
Does Pakistan have an equivalent of the Digital Markets Act?
Pakistan does not presently operate the same EU-style gatekeeper regime. However, section 3 of the Competition Act 2010 prohibits abuse of dominant position, including exclusionary and discriminatory practices capable of preventing, restricting, reducing or distorting competition.
Could a technology platform be dominant under Pakistani law even with less than 40 per cent market share?
Potentially, yes. The CCP explains that while a market share above 40 per cent creates a statutory presumption of dominance, dominance may also be established below that threshold where an undertaking can behave to an appreciable extent independently of competitive constraints.
Why should Pakistani AI companies care about European competition regulation?
Because Pakistani technology businesses may supply products into Europe or depend upon global operating systems, search engines, app stores and AI ecosystems affected by European regulation. European rules concerning interoperability and platform access can therefore influence the commercial opportunities available to companies incorporated outside Europe.
The wider legal lesson: monopoly power has changed shape
The twentieth-century monopolist might have owned the factory.
The twenty-first-century gatekeeper may own something less visible.
The operating system beneath the factory.
The dataset telling customers which factory to choose.
The app store through which the factory reaches them.
The algorithm ranking the factory.
And increasingly, the artificial intelligence assistant which decides whether the customer needs a factory at all.
That is why competition law is moving into territory once regarded as purely technical.
Wake words.
APIs.
Search data.
Default assistants.
Context permissions.
Ranking systems.
On-device models.
These may sound trivial compared with railways and oil pipelines.
They are not.
Digital civilisation increasingly moves through them.
The future antitrust case may begin with a single sentence
“Hey, AI, buy this for me.”
The user may never know which businesses were considered.
Which were excluded.
Which were given preferential treatment.
Which platform supplied the data.
Which company paid for prominence.
Which application was invoked.
Which competitor could not participate because it lacked access to the operating system.
Convenience will conceal architecture.
Architecture will conceal power.
And competition law will increasingly have to look beneath the answer.
The European Commission’s July 2026 intervention therefore represents something larger than another instalment in Brussels’ long-running disputes with Google.
It suggests a fundamental change in what competition regulation regards as economically significant.
The critical resource may no longer be the product.
It may be the interface through which products become visible.
The critical monopoly may not concern what consumers purchase.
It may concern who is permitted to recommend the purchase.
And the critical gate may not stand at the entrance to a railway, port or marketplace.
It may sit quietly inside the operating system of the telephone in your hand.
Europe has now decided that, at least for designated gatekeepers, some of those gates must open.
Whether that produces greater innovation or excessive regulatory intervention will be tested in the years ahead.
But the jurisprudential direction is already clear.
The next generation of competition law will not merely ask who dominates a market.
It will ask:
Who controls the architecture through which everyone else must compete?
For the age of artificial intelligence, that may be the more important question.
About the Author
Barrister Aemen Zulfikar Maluka is the founder of Josh and Mak International, an Islamabad-based law practice with an international commercial and regulatory advisory outlook.
Her work examines the increasingly porous boundary between Pakistani and international law, particularly where foreign regulatory developments affect technology companies, investors, exporters, digital businesses and cross-border commercial relationships connected with Pakistan.
Rather than treating European, British, American or Asian regulatory developments as remote academic events, Barrister Aemen’s international commentary focuses upon their practical consequences: who gains market access, which contracts need to change, what new liabilities arise, how businesses should structure themselves and where emerging foreign legal rules may foreshadow regulatory questions Pakistan will eventually confront itself.
Competition law in the technology era is a particularly good example. A Pakistani business may never establish a European subsidiary and yet remain commercially affected by European rules governing the global operating systems, search engines, digital marketplaces and AI infrastructure upon which its business depends.
For further insights, international competition and regulatory analysis, cross-border technology advice or assistance concerning the implications of global digital regulation for businesses operating from or connected with Pakistan, contact Barrister Aemen at Aemen@joshandmak.com.
Josh and Mak International
www.joshandmakinternational.com
This article is intended as general legal and regulatory commentary. It does not constitute jurisdiction-specific legal advice, and application of competition or digital-platform regulation should be considered by reference to the relevant undertaking, market, conduct and jurisdiction.
Filed Under : EU AI competition law 2026, Google AI antitrust 2026, Google Digital Markets Act fine, Google €890 million fine, EU Google DMA 2026, AI interoperability law, Google Android AI interoperability, Gemini competition law, competing AI assistants Android, Article 6(7) DMA, Digital Markets Act interoperability, Android interoperability rules, Google Search data sharing, Article 6(11) DMA, AI chatbot search data, search data competition law, ranking query click view data, FRAND search data, European Commission Google July 2026, Google self-preferencing 2026, Google Play anti-steering, platform self-preferencing law, AI gatekeeper regulation, AI antitrust Europe, artificial intelligence competition law, AI assistants competition, default AI assistant antitrust, AI operating system competition, AI agents competition law, digital gatekeeper law, Big Tech competition regulation, technology monopoly law, data monopoly competition, data network effects antitrust, platform interoperability regulation, API competition law, AI API access, digital ecosystem competition, algorithmic gatekeeping, AI recommendation competition, AI search antitrust, generative AI competition regulation, EU Digital Markets Act Pakistan, competition law Pakistan digital platforms, Competition Act 2010 section 3, abuse of dominant position Pakistan, digital market dominance Pakistan, CCP digital markets, Competition Commission Pakistan technology, AI competition law Pakistan, Pakistani AI companies Europe, Pakistan technology exporters EU law, international competition lawyer Pakistan, cross-border technology law Pakistan, global AI regulation, AI market access law, digital platform regulation Pakistan, app store competition law, operating system antitrust, AI data access rights, AI interoperability rights, platform access competition, AI commercial contracts, AI regulatory due diligence, technology merger due diligence, AI startup competition law, artificial intelligence market power, AI ecosystem law, AI agents market power, competition law default settings, AI recommendation algorithms law.