1. An enabling and protective architecture

The DACC are a doctrinal construction intended to serve as a model for national, European and international legislators, as guidance for designers and developers of digital assistants for consumer use and as a source of inspiration for self-regulation and standardisation (Art. 1(2) DACC).Footnote ⁸

The DACC propose that consumers should have the right to use a digital assistant in their contractual relations with a business (Art. 3(1)). Businesses other than micro or small enterprises must not contractually restrict that right, or design, organise or operate their online interfaces in a manner that prevents consumers from using digital assistants (Arts. 3(2)–(3) and (5)).

The proposed right is embedded in an extensive protective architecture combining design requirements for digital assistants and obligations imposed on the suppliers of digital assistants. Digital assistants must be designed so as to allow consumers to select and modify parameters to be used by the digital assistant in performing its functions (Art. 5), choose between confirmation and objection mechanisms (Art. 6), and deactivate the digital assistant temporarily or permanently (Art. 7). They must include a functionality for disclosure that a digital assistant is used (Art. 8), they must not be designed or operated in a way that manipulates a consumer or otherwise materially distorts or impairs the ability of a consumer to make free and informed choices or decisions (Art. 9) and they must document how a particular decision is made (Art. 10). The design requirements are meant to provide consumers with ex ante protection: only digital assistants that comply with these requirements should be supplied to consumers (Art. 4(1)). The DACC leave it to the legislator to determine the process that needs to be followed to reach that aim.Footnote ⁹ The comments on Article 4 add a troublesome specification concerning the scope of the design requirements. Digital assistants offered for free, meaning “without anything given by the consumer in return, such as payment or access to data (…), other than those integrated into a smart product, and open-source digital assistants, are not the focus of those requirements.” Those requirements could, however, serve “as a general standard for digital assistants and (…) as a guide also for developers of free/open-source digital assistants.”Footnote ¹⁰

In the relationship between the consumer and the digital assistant’s supplier, the DACC impose information obligations (Art. 11), require disclosure of conflicts of interest (Art. 12), regulate deactivation of the digital assistant (Art. 13), conformity and liability (Art. 14 and 16) and require warnings when the supplier possesses clear and reliable information that a business regularly infringes consumer law or contractual obligations and has not taken adequate steps to prevent such infringements in the future (Art. 15). Article 17 provides that an algorithmic contract should not be denied validity or enforceability solely because a digital assistant was used, while Article 18 confirms that using a digital assistant does not deprive the consumer of the protection of consumer law.

Further rules govern the assistant’s interaction with traders. They concern disclosure that a digital assistant is being used (Art. 19), oblige businesses other than micro or small enterprises to provide legally required information in machine-readable form (Art. 20), and allow information duties to be discharged through the digital assistant under specified conditions (Art. 21). Article 23 prohibits businesses from designing or operating interfaces in a way that is likely to materially distort or impair a digital assistant’s functioning and permits consumers to set aside contracts resulting from such manipulation.

The attribution regime combines a general rule attributing the digital assistant’s conduct to the consumer with an exception for actions departing from what the consumer could reasonably expect (Art. 22). When the digital assistant’s actions are not attributed to the consumer, its supplier must compensate the affected business for the resulting loss (Art. 25).

In sum, the DACC combine an enabling entitlement with ex ante design requirements, contractual protections and ex post liability rules.Footnote ¹¹ They seek to increase legal certainty and provide a high level of consumer protection in business-to-consumer contracts where one or both parties use a digital assistant and to allocate the risks arising when delegation produces unexpected outcomes.Footnote ¹² They do not aim to provide a self-standing regime for digital assistants; they are designed to interact with other rules, for example on consumer contracts, data protection and cybersecurity (Art. 1(4)).

2. Timely regulation and the question of sequencing

According to Busch, one of the DACC co-reporters, consumer law should enable the use of agents where they can empower consumers, while ensuring adequate protection when agents are deployed.Footnote ¹³ His more recent account of “agent-ready” consumer law favours incremental but timely adaptation. It identifies the validity and enforceability of agentic contracts, protection against manipulation of agents and agent-compatible acknowledgement of payment obligations as foundational issues.Footnote ¹⁴ It also envisages clarification of the application of the existing legal framework to agentic commerce as well as the translation of human-facing consumer protections to the protocol layer as the technical arrangements governing agentic transactions become clearer.Footnote ¹⁵ Busch warns that delayed intervention may allow harmful practices and market structures to become entrenched and concludes that, at present, the risk of intervening too late outweighs that of intervening too early.Footnote ¹⁶

At present, deployment in the EU remains limited.Footnote ¹⁷ However, CAIAs such as Perplexity’s Comet have already given rise to litigation in the US.Footnote ¹⁸ The case for preparing consumer law for agentic commerce is therefore compelling. CAIAs may reduce search costs, process information that would overwhelm human consumers, identify contractual restrictions and, unless they are trader-specific or restricted to “walled gardens,”Footnote ¹⁹ compare products across a wider range of traders.Footnote ²⁰

Legislative inaction would leave the determination of the conditions for agent-mediated access and significant elements of consumer protection to a significant extent to private ordering. This raises concerns of democratic legitimacy, because these issues do not only concern the businesses setting the rules, but also consumers and other market participants. If legislative intervention occurs too late, market structures, contractual practices and technical arrangements may have become difficult to change.Footnote ²¹ This article therefore agrees with Busch that regulation should begin before agentic markets become entrenched and should be gradual. The more specific question examined here is where, within that regulatory sequence, a general DACC-style right to use a digital assistant defined in such a way that it includes CAIAs, should be recognised.

1. The positive right and its expressive effect

A pure no-barrier provisionFootnote ²² would prevent traders from discriminating against agent access or using interface control to foreclose agent access without objective justification. The law can also adopt rules clarifying the validity and attribution of CAIA-mediated transactions, preserving consumer protection when a CAIA is used, and protecting users of CAIAs against manipulation. A right to use a CAIA communicates more. The expressive function of the law helps explain why the practical meaning of such a provision may exceed its doctrinal content.Footnote ²³

The recognition of a right to use a CAIA presents agent-mediated contracting as a legitimate form of market participation that businesses must ordinarily accommodate. Statutory recognition reassures consumers, normalises the protected activity and encourages adoption. The DACC’s surrounding protective rules would reinforce that signal. The legal order would not merely tolerate consumers’ use of unregulated tools at their own risk; it would prescribe control mechanisms, regulate attribution and provide remedies. Taken together, these provisions could convey that consumers are entitled to use CAIAs, that CAIAs complying with the applicable requirements are reliable, and that the law offers protection when something goes wrong. This provides a stronger basis for consumer confidence and adoption than a prohibition on obstruction.

From the perspective of consumer protection, the main risk would be that consumers overestimate the protection the law offers and, in particular, that they confuse the existence of a regulatory framework with the technical capabilities of CAIAs.Footnote ²⁴ This is explained further in the next section.

2. The safety halo as an overextension of justified institutional trust

If translated into binding and adequately enforced rules, the DACC would justify trust in the safeguards they expressly establish: consumers could, for example, reasonably trust that they can specify and revise the digital assistant’s operating parameters (Art. 5), that the provider discloses conflicts of interest (Art. 12), that certain unexpected actions fall outside attribution (Art. 22), and that non-conformity, including incorrect operation of the functionalities prescribed by the design requirements, leads to supplier liability (Art. 16 juncto Art. 14). The Commentary on Article 4, however, casts doubt on whether the translation of the design requirements into binding law should extend to genuinely free standalone and open-source assistants, for which it suggests that they may instead serve only as a general standard or guide, as discussed in Section II.1. Users of those assistants would therefore lack an equivalent basis for justified trust in the design safeguards. This limitation is particularly concerning because those are the assistants most likely to be used by economically vulnerable consumers.

Even where the design requirements are binding, consumers should not infer from the DACC’s right to use a digital assistant that the CAIA is technically reliable, understands every broad or ambiguous instruction correctly, identifies the objectively best product, or resists adversarial attacks. Recent computer science studies conducted in controlled experimental environments reveal substantial weaknesses in end-to-end task completion, preference interpretation and the verification of product suitability. In ShoppingBench, for example, an agent based on GPT-4.1 successfully completed fewer than half of the tested tasks, which included budget management, voucher use and finding a seller offering all regulated products.Footnote ²⁵ In ShopperBench, an agent based on Claude Haiku 4.5 was supplied with profiles describing general consumer preferences. It added an acceptable product to the shopping cart in only about 63% of tasks, even where the profile had been translated into specific constraints such as a maximum price.Footnote ²⁶ WebMall tested agents across four simulated webshops on tasks ranging from product search and comparison to adding products to a cart and completing checkout. Even the strongest configurations achieved task completion rates below 65% in the task groups cheapest product search and vague product search.Footnote ²⁷ All three benchmarks use constructed tasks in controlled or simulated environments. The task-completion figures therefore do not establish error rates in actual consumer transactions and may encompass non-completion as well as incorrect execution. The studies nevertheless provide complementary evidence of substantial limitations even under controlled conditions.

Current agents also appear to be vulnerable to features of the commercial and technical environment that are unrelated to the consumer’s preferences. Computer science research shows significant and model-dependent position effects, sensitivity to presentation and concentration of demand on a limited number of products. In some cases, changes to product descriptions led to substantial increases in the share of products selected by buyer agents.Footnote ²⁸ Furthermore, web agents appear susceptible to instructions embedded in third-party content: in the WASP benchmark study, agents began following low-effort adversarial prompt injections in up to 86% of cases, although their limited execution capabilities often prevented completion of the attacker’s full objective.Footnote ²⁹

Present systems also do not reliably minimise their use of personal data. A study found that agents built on several leading model families were often prone to using unnecessary sensitive information while performing web tasks.Footnote ³⁰

The force of these concerns, nevertheless, varies with the object and scope of the task delegated to a CAIA. A CAIA that exercises discretion only in interpreting a consumer’s preferences among offers within a particular trader’s environment undertakes a more delimited task than a CAIA asked to identify the product that best fits a broadly stated objective across multiple traders and sources of information. Confining the CAIA’s task may limit the range of possible errors, but it does not establish that it can reliably exercise its remaining discretion when the consumer’s preferences are incomplete, implicit or must be inferred from a profile or elicited through dialogue. This is illustrated by EComAgentBench, in which performance was lower when relevant requirements had to be recovered from a profile or elicited through clarification rather than read from the initial instruction.Footnote ³¹ Although the benchmark used synthetic profiles and a restricted product set, it illustrates that narrowing the scope of a CAIA’s task does not itself establish reliable preference interpretation.

These findings indicate that current agents may act inconsistently with the consumer’s instructions, expectations or interests. The DACC do not certify that CAIAs can reliably interpret and apply broad consumer intentions, verify the information relevant to their decisions or operate robustly in adversarial commercial environments. Legal recognition may nevertheless be understood as a more general judgment that agent-mediated contracting has become technically reliable. The distinction between justified trust in the legal framework and trust in the agent’s substantive competence is therefore critical. An unjustified “safety halo” may arise where the recognition of a right to use a CAIA not only evokes trust in the legal framework, but is perceived as an institutional recognition of the technical reliability of the technology.

3. Formal protection and the practical limits of redress

The DACC mitigate several of the risks identified in the previous section. Parameters, confirmation and objection mechanisms give consumers opportunities to constrain or interrupt the assistant’s actions (Arts. 5–6). Disclosure of conflicts addresses some sources of distorted selections (Art. 8), while the prohibition of manipulation applies both to the operation of the assistant and to trader interfaces (Arts. 9 and 23). Decision documentation may help reconstruct why a particular product was selected (Art. 10). Where the assistant does not conform to the supply contract, the consumer may claim against its supplier, including for resulting losses (Arts. 14 and 16). In relation to the trader, an action departing from what the consumer could reasonably expect has no legal effect and is not attributed to the consumer (Art. 22), while a contract resulting from manipulation by the trader may be set aside (Art. 23(2)).

Yet, these rules do not guarantee that the CAIA will accurately interpret the consumer’s instructions, identify the option that corresponds best to the consumer’s desires or resist adversarial influences. A CAIA may misunderstand a broad instruction, fail to identify relevant information, allocate a budget poorly or be influenced by the presentation of offers. Such an outcome does not necessarily make the assistant non-conforming within the meaning of Article 14, place its action outside the consumer’s reasonable expectations within the meaning of Article 22Footnote ³² or establish manipulation by the trader within the meaning of Article 23.Footnote ³³ A consumer who confirms a flawed recommendation or who chooses the objection model and does not intervene in time will therefore ordinarily remain bound under the DACC. This does not necessarily make the transaction irreversible. Since the CAIA-mediated transaction qualifies as a distance contract, Article 9 Consumer Rights Directive generally allows the consumer a fourteen-day period to withdraw from the contract without giving a reason.Footnote ³⁴ Its protection remains limited, however. The consumer must discover the problem and communicate the decision to withdraw within the applicable period, bears the burden of proving that withdrawal was exercised,Footnote ³⁵ and may incur return costs.Footnote ³⁶ The withdrawal right is also subject to exceptions.Footnote ³⁷ Moreover, withdrawal cannot reverse personal data disclosure, missed opportunities or reputational harm.

Moreover, formal entitlement to a remedy does not ensure effective redress. Although Article 22 DACC is formulated as a rule of attribution and legal effect, the consumer may still have to contest the trader’s demand for performance or seek restitution of performance already rendered. Setting aside a contract under Article 23 requires the consumer to identify the trader’s manipulation and establish the connection with the resulting contract. A claim against the supplier under Article 16 requires the consumer to establish non-conformity, loss, causation and any further conditions imposed by the applicable law. The consumer must detect the problem, determine whether the unwanted outcome resulted from the assistant’s operation, manipulation through the trader’s interface, erroneous or unavailable product or price data, or malicious third-party interference. A single failed transaction may require the consumer both to resist enforcement by the trader and to seek compensation from the CAIA supplier.

Article 10 documentation can assist, but it does not eliminate these obstacles. The evidence may be technically difficult to interpret or divided among the supplier, the trader and third-party data providers. Even if implementing EU legislation required courts to apply relevant consumer protections of their own motion, that safeguard would operate only once the dispute had reached a court. It would not remove the consumer’s initial burdens of detecting the problem, challenging the outcome and securing the necessary evidence. Consumers may rationally decide not to contest an undesired CAIA-mediated outcome where the expected benefits of redress do not justify the costs, effort and uncertainty involved.Footnote ³⁸ Consequently, losses for which a remedy formally exists may in practice remain with the consumer. Where a design or operational deficiency affects many consumers, representative actions can reduce this enforcement deficit by enabling qualified entities to seek injunctions and redress on behalf of groups of consumers. Representative actions should be complemented by public enforcement because systemic failures may require the investigative and coordinated enforcement powers available to competent authorities under the Consumer Protection Cooperation Regulation.Footnote ³⁹ Any EU instrument implementing DACC-inspired rules should therefore be added to Annex I to the Collective Redress DirectiveFootnote ⁴⁰ and to the Annex to the Consumer Protection Cooperation Regulation, so that qualified entities and competent authorities can address infringements affecting groups of consumers. Collective and public enforcement may improve deterrence and compensation and put an end to illegal conduct, but they cannot ensure correct performance of all delegated tasks or undo consequences such as the disclosure of confidential data.Footnote ⁴¹

More demanding ex ante design requirements could reduce the incidence of unwanted outcomes and, where effective, would protect consumers better than remedies available only afterward. For certain uses, legislation could require providers to ensure that agents observe stated constraints, seek clarification of material ambiguity and verify decisive transaction information, while harmonised standards could specify methods for assessing compliance. Nevertheless, conformity under defined test conditions would not establish that a CAIA reliably interprets incompletely specified objectives across changing transactional environments. Moreover, it is questionable whether the current state of the technology enables the deployment of CAIAs that can consistently satisfy demanding requirements of this kind. The possibility of strengthening the ex ante framework therefore does not justify the immediate recognition of a general right.

1. Legal recognition as a market-constituting force

The recognition of a right to use a digital assistant, insofar as that right encompasses CAIAs, does not only communicate to consumers but also to other groups whose responses shape the market. For CAIA providers, the regime indicates that consumers will be legally entitled to use their products and that traders will face limits on excluding them. For traders, it signals that agent-mediated traffic may become an increasingly important route to consumers. For standard setters, identity providers and payment services, it indicates growing demand for protocols governing consumer instructions, authentication, revocation and machine-readable contracting. Legal recognition can consequently accelerate market development even where the legislator’s stated purpose is consumer protection rather than technological promotion.

Law and political economy scholarship shows that markets are constituted through legal entitlements and obligations and through the institutional arrangements that enforce them.Footnote ⁴² In digital markets, legal rules governing access, appropriation, contracts and immunity have contributed to the emergence and consolidation of platform power.Footnote ⁴³ The recognition of a right to use a CAIA may similarly help constitute the market in which CAIAs operate. The relevant regulatory question is therefore not simply whether the right stimulates innovation; it is which forms of innovation and dependency the right facilitates.

Busch recognises that the manner in which agents are deployed matters for market structure. He distinguishes marketplace-embedded agents, independent third-party agents and agents integrated into operating systems, and identifies OS-level integration as raising particularly acute gatekeeping concerns.Footnote ⁴⁴ The present argument develops the structural implications of that observation: a positive right that accelerates adoption may intensify existing asymmetries by increasing consumers’ willingness to adopt whichever CAIAs are available first and appear most institutionally established.

2. Disproportionate benefit for incumbents in adjacent digital markets

A positive right to use a digital assistant that encompasses CAIAs may be formally neutral while disproportionately benefiting incumbents in adjacent digital markets. They may be able to respond especially quickly because they are already active across several complementary layers on which CAIAs depend, including cloud computing, data, foundation models, operating systems and distribution channels, as well as downstream services such as search and online marketplaces. An operating-system provider can integrate an agent into the device. A browser provider can make its agent the default route to websites. A marketplace can combine its agent with direct access to inventory, payments, fulfilment and complaints mechanisms. A search provider can integrate agent functions with existing access to commercial information.Footnote ⁴⁵ Independent providers, by contrast, need additional time to construct the technical, commercial and institutional relationships necessary to offer a credible alternative.

This temporal disadvantage matters because the period immediately following legal recognition may be unusually important. The right and its accompanying remedies may increase consumer willingness to use CAIAs and induce traders to invest in CAIA-compatible interfaces. Early CAIAs deployed by incumbent market players may begin accumulating users and integrating with traders while independent alternatives are still being developed. These early choices can make later entry more difficult even where no provider was formally excluded. The concern is therefore that regulatory recognition may narrow the window in which independent providers can contribute to shaping the market.

The protective requirements necessary to make the right credibleFootnote ⁴⁶ can contribute to this dynamic. Research on data-intensive markets shows that fixed regulatory and transaction costs may burden small and new firms more heavily than incumbents and may reduce investment in emerging technology ventures.Footnote ⁴⁷ Although these studies concern privacy regulation rather than CAIAs, they support the general mechanism by which formally uniform protective requirements may increase comparative entry delays. An exemption for small CAIA providers is not a satisfactory solution: it would deprive their users of core protections, and it could reinforce the perception that smaller providers are less trustworthy. An unfortunate interaction thus arises: requirements intended to reduce the risks of legally recognised agentic delegation may also favour incumbents in adjacent markets entering the CAIA market, which can absorb the associated compliance costs more readily than independent CAIA providers. Where an early CAIA is deployed by an incumbent controlling an online marketplace, a two-way feedback loop occurs: Direct access to product data, checkout, delivery information and redress can improve the affiliated CAIA. The CAIA can in turn strengthen the marketplace by steering consumer demand towards affiliated products, preferred sellers or technically integrated offers. Marketplace integration thus makes the CAIA more useful, while control over the CAIA increases the marketplace’s power over the route through which traders reach consumers.

Where a CAIA is used repeatedly, it may become increasingly valuable as it accumulates preferences, transaction histories, account connections and knowledge of the consumer’s reactions to earlier selections. This makes the initial selection of a CAIA especially important. Once the CAIA has built a detailed representation of the consumer, the option to switch to another CAIA will no longer provide effective contestability. Switching requires not merely installing another application, but transferring or reconstructing the consumer profile and relevant history and reconfiguring the parameters and account integrations on which the CAIA’s operation depends. Article 5(4) DACC partially addresses the informational dimension of switching by requiring access to a current or historic record of the parameters and the option to receive a copy on a durable medium. It does not, however, require that the parameters be provided in a machine-readable or interoperable format, and its Commentary expressly leaves portability between digital assistants for later reconsideration.Footnote ⁴⁸ Article 20 GDPR provides for the portability of personal data, but not for the functional portability of the agent relationship. Even where the consumer obtains a machine-readable file, the new CAIA may use different categories, formats and inference methods and may be unable to reproduce the old CAIA’s personalisation.Footnote ⁴⁹

These advantages are amplified by network effects. Agranat and Gal argue that when many consumers entrust their purchasing decisions to the same agent, the agent aggregates their previously separate buying decisions and thereby acquires greater bargaining power over traders, resulting in better prices or other favourable terms for its users, which will in turn make the service more attractive to additional consumers. Their arrival further increases the agent’s bargaining power, creating a self-reinforcing feedback loop. This process may produce a small number of powerful agent hubs, whose performance increasingly reflects their network size rather than superior technology.Footnote ⁵⁰

The ability to enter quickly after legal recognition may therefore provide more than a temporary first-mover advantage. Incumbent gatekeepers may be especially well placed to provide the CAIAs to which the first substantial groups of consumers attach. Once their initial user bases generate greater bargaining leverage and better transactional outcomes, later entrants may struggle to attract users even where their technology or alignment with consumer interests is superior. A positive right that accelerates adoption may thus increase not only the speed of market development but also the likelihood that the market tips towards the providers able to deploy at scale from the outset.

3. SME invisibility and dependence on marketplaces

The transition to agentic commerce may affect traders unevenly. CAIAs operate more effectively where product and transactional information is structured, machine-readable and technically accessible. Structured product data will become important as consumers increasingly reach traders through CAIAs rather than through human-facing websites.Footnote ⁵¹ However, supplying and maintaining current product information, prices, inventory data and machine-accessible transactional functions requires data infrastructure, technical skills and organisational investment. Research on retail digitalisation shows that smaller retailers face barriers to adopting digital technologies, such as limited budgets, low levels of digital literacy, and inadequate infrastructure.Footnote ⁵² They often cannot pursue integrated digital-channel strategies at the same scale and speed as larger firms.Footnote ⁵³ Although the evidence does not directly concern agentic-commerce markets, it is plausible that large retailers and marketplaces will become CAIA-ready more rapidly than many SMEs.

The DACC exempt micro and small enterprises from important aspects of the no-barrier principle: the prohibition on contractual restrictions and interface practices that prevent consumers’ use of digital assistants (Arts. 3(2)–(3)), and the duty to provide required contractual information in a structured, commonly used and machine-readable format (Art. 20). This reflects legitimate proportionality concerns. Yet, exemption from legal duties does not ensure commercial visibility, or, more precisely, commercial legibility. Cohen uses “legibility” to describe rubrics through which intermediaries render users knowable and actionable; such rubrics encode particular ways of understanding and managing them.Footnote ⁵⁴ Extending Cohen’s concept to agentic commerce, machine readability can be understood as a component of trader legibility. SMEs may therefore remain formally accessible while being insufficiently legible to CAIAs.

Marketplaces or data intermediaries can remedy that deficit by translating SME offerings into common machine-readable structures, but in doing so they may acquire power over the classifications and informational conditions through which SMEs reach CAIA-mediated demand. SMEs may therefore become dependent on marketplaces or data intermediaries that aggregate, structure and classify their offerings.

4. Private technical governance, standardisation and legitimacy

Even if the EU legislator enacted rules corresponding to the DACC in full, the resulting legislation would not exhaustively determine how CAIAs act or interact in agentic commerce. It would not, for example, determine how a CAIA must proceed where the consumer’s instructions are ambiguous or incomplete or how it must resolve any conflict that remains after taking account of the relative importance assigned to different parameters.Footnote ⁵⁵ Nor would it prescribe a shared technical architecture for interactions among CAIAs, traders and other service providers. Although Article 20 requires contractual information to be provided in a structured, commonly used and machine-readable format, the DACC do not prescribe the data schemas or interfaces through which that information must be supplied. Nor do they prescribe the technical mechanisms through which CAIAs are authenticated, the scope of the consumer’s instructions is communicated, or orders, payment requests and transaction records are transmitted and recorded.Footnote ⁵⁶ They also do not require interoperability between CAIAs and trader systems or require parameters and profiles to be transferable in a form that another CAIA can use.Footnote ⁵⁷ In the absence of supplementary public rules, these matters would be determined in practice by individual market players, industry consortia or standardisation organisations.

Private determination of the conditions under which CAIAs can interact with traders and other service providers may affect the contestability of agentic markets. Research on competition for technological dominance shows that early entry can produce a lasting advantage where network effects are present.Footnote ⁵⁸ A CAIA provider with a substantial user base may induce traders and payment providers to make their systems compatible with its CAIA. Those integrations make the CAIA more useful to consumers and may attract further users.Footnote ⁵⁹ Competing CAIAs may then need to satisfy the technical requirements already adopted by traders and service providers to obtain comparable access. Those requirements may consequently become de facto standards and conditions of effective market participation. Legal recognition of a general right to use a digital assistant, although formally technology-neutral, could reinforce this process by accelerating the adoption of early CAIAs.

Apart from their possible effects on market structure, the choices left to private actors raise a question of legitimacy. Technical implementation necessarily entails private design decisions. Different implementations may comply with the express requirements of the DACC while making consumer intervention more or less effective, permitting different degrees of reliance on inferred preferences or imposing different technical burdens on CAIA providers and traders. Private design choices may therefore perform a rule-setting function: they determine how legal requirements operate and how questions on which the law is silent are resolved in practice.Footnote ⁶⁰ This raises concerns of democratic legitimacy.

In the EU, detailed technical implementation is likely to be channeled through European standardisation pursuant to a Commission standardisation request under the New Legislative Framework.Footnote ⁶¹ This mechanism originates in the field of product regulation, but it has recently been extended to AI regulation.Footnote ⁶² Compliance with harmonised standards developed under this mechanism remains voluntary, but once their references have been published in the Official Journal of the European Union, conformity with those standards gives rise to a presumption of conformity with the legal requirements they cover.

The procedure for setting the harmonised European standards is, however, itself fraught with issues of legitimacy as a result of a lack of transparency and participation in the standard-setting process.Footnote ⁶³ These deficiencies strengthen the case for broader reform of the European standardisation system.Footnote ⁶⁴ They do not, however, necessarily mean that the development of standards for CAIAs should await the completion of that reform. Pending broader institutional reform, the legitimacy risks should therefore be mitigated within the existing framework. The legislator should determine the essential choices, for example concerning the scope of delegation and the treatment of ambiguous instructions. The Commission’s standardisation request should delimit the remaining technical task with sufficient precision, while the Commission and the European standardisation organisations should make full use of the possibilities for transparency and balanced stakeholder participation available under the existing framework.Footnote ⁶⁵

Because harmonised standards are formally part of EU law,Footnote ⁶⁶ conformity may be understood by market participants as an official assurance of a CAIA’s general reliability, thereby contributing to the safety halo. Should this mechanism be used, it should therefore be communicated very clearly that any presumption of conformity is strictly limited to the specific legal requirements covered by the standard. It should not be taken to imply that the CAIA will correctly interpret the consumer’s objectives, resolve ambiguity in the consumer’s interests or produce a substantively satisfactory outcome.

5. Deterioration of non-agent participation

The DACC right to use a digital assistant protects consumers who choose to delegate through a digital assistant, including a CAIA. However, it may also contribute to a market transformation that disadvantages consumers who do not delegate. Traders may devote increasing resources to CAIA-facing access and reduce investment in human-facing interfaces. Better prices, availability or customer-service functions may become easier to obtain through CAIAs. Human consumers may technically retain access but receive a slower or materially inferior route. A rapid transition to agentic commerce could therefore make CAIA use practically advantageous or even difficult to avoid without formally requiring it. The recognition of one form of autonomy, the freedom to delegate, could then weaken another: the effective freedom to participate without delegation.

The DACC partly anticipate the danger that consumers may be compelled to delegate. Article 3(4) prohibits businesses from requiring consumers to use their own (i.e., the consumers’) digital assistants and protects the main functionality of products where the use of an integrated assistant is declined. Article 7(3) provides equivalent protection upon deactivation. These provisions preserve an important element of residual human control. Their scope is nevertheless limited. The commentary states that Article 3(4) does not extend to an algorithmic system forming an integral part of a trader’s website, and the DACC do not expressly require human-facing channels to offer equivalent prices, availability, functionality or service quality.

6. The DMA and EU competition law as partial responses

The structural risks identified above do not arise in a regulatory vacuum. The DMAFootnote ⁶⁷ can constrain certain forms of incumbent leveraging through ex ante obligations imposed on designated gatekeepers in relation to designated core platform services. The DMA applies according to the function performed by a service rather than its use of AI. Where a gatekeeper incorporates a CAIA or other AI functionality into a designated core platform service, the obligations governing that service apply. The Commission’s first review of the DMA confirms that several DMA obligations already fully apply to certain forms of AI technology when they are integrated in designated services. It refers to default settings under Article 6(3), interoperability obligations under Article 6(7) and access to search data under Article 6(11).Footnote ⁶⁸ In the meantime, the Commission adopted a decision specifying the measures Alphabet must put in place to comply with its interoperability obligations under Article 6(7) DMA in relation to certain Google Android features relevant for the provision of AI services.Footnote ⁶⁹ The decision illustrates how the DMA can prevent an operating-system gatekeeper from reserving capabilities needed by AI services, including CAIAs, to its own integrated service. The Commission further considers the possible designation of certain AI services as virtual assistants, and possible use of an Article 19 DMA market investigation to consider whether AI services should be added as a new core platform service or whether existing obligations require amendment.Footnote ⁷⁰ Bostoen and Krämer agree that these two mechanisms enable the DMA to capture CAIAs. However, the application of the DMA to a CAIA as a standalone service remains dependent on its designation as an (emergent) gatekeeper which requires at least a foreseeably entrenched and durable position in the near future.Footnote ⁷¹ Moreover, like the Commission,Footnote ⁷² they acknowledge that the DMA cannot capture all competition issues in the AI value chain.Footnote ⁷³

Article 102 TFEU catches types of abuse for which no specific DMA obligation is available or which are committed by non-designated undertakings. When an undertaking is dominant, restrictions on interoperability, discriminatory access conditions, tying, self-preferencing, exclusivity or foreclosure may constitute an abuse. For example, where a dominant provider controls the technical interface or access rules through which CAIAs interact with its service, discriminatory access conditions, exclusionary design or the leveraging of that control into adjacent markets may amount to an abuse of dominance.Footnote ⁷⁴ In Google Shopping, the Court of Justice confirmed that a dominant undertaking which treats its own products or services more favourably than it treats those of its competitors may under certain circumstances infringe Article 102 TFEU.Footnote ⁷⁵ In Android Auto, the Court held that a dominant undertaking’s refusal to ensure that a digital platform, not developed solely for its own business needs, is interoperable with third-party apps may constitute an abuse of dominance, even if that platform is not indispensable to the commercial operation of the apps in a downstream market.Footnote ⁷⁶ These cases show that Article 102 TFEU may prohibit exclusionary control over an interface or infrastructure even where no corresponding DMA obligation applies. Recently, the Commission imposed interim measures requiring Meta to preserve access to WhatsApp for competing AI assistants pending an investigation of potential abusive refusal of access.Footnote ⁷⁷ Equally, it opened an investigation into Google’s possible imposition of unfair terms on publishers whose content is used for AI Overviews and AI services.Footnote ⁷⁸ These proceedings do not establish a final infringement, but they demonstrate that competition law can reach AI-related conduct involving access, data and ecosystem integration.

Article 101 TFEU is also relevant to the formation of agentic consumer markets. Many of the technical standards and protocols governing agent authentication, the communication of consumer instructions, trader access, payments and transaction records may be developed collectively by industry participants, consortia or standard-setting bodies. Agreements on common technical standards or protocols may facilitate interoperability and market entry, but they may also exclude competing technologies, discriminate between participants or facilitate coordination. Article 101 TFEU can scrutinise such arrangements where they have the object or effect of restricting competition.

EU merger controlFootnote ⁷⁹ may intervene before exclusionary conduct occurs where an acquisition or structural partnership constitutes a concentration and threatens to consolidate control over CAIA services or distribution channels. Its reach is nevertheless incomplete. Partnerships that do not confer lasting control fall outside Article 3 EUMR.Footnote ⁸⁰ Acquisitions of nascent AI firms may not have a Union dimension under the Article 1 turnover thresholds.Footnote ⁸¹ Such transactions are therefore not systematically subject to ex ante Commission merger review, although some may be reviewed under national merger control rules or referred to the Commission under Article 22 EUMR.Footnote ⁸²

Therefore, those instruments do not remove all of the earlier structural concerns. First, they do not eliminate the risk that patterns of consumer adoption and trader integration around particular technical standards or interfaces become entrenched before the thresholds for dominance or gatekeeper designation are met. Second, they do not ensure SME legibility. The DACC’s limited machine-readability duty would not fill that gap for micro and small enterprises. They may be formally reachable but practically difficult for CAIAs to identify, compare or transact with. Third, those instruments do not determine how privately developed technical standards and access rules governing CAIA–trader interactions should be established or what those arrangements should contain. Fourth, the DMA and competition law do not require functionally equivalent human-facing channels where traders increasingly optimise prices, availability or service for CAIA-mediated transactions.

Moreover, the timing and evidential structure of competition law do not match all market formation risks. Article 102 TFEU requires proof of a dominant position within the internal market or a substantial part of it and of an abuse of that position. Article 101(1) TFEU requires proof of an agreement between undertakings, a decision by an association of undertakings or a concerted practice that has an anticompetitive object or actual or potential appreciable restrictive effects on competition within the internal market. In both cases, the conduct must be shown to be capable of appreciably affecting trade between Member States. Furthermore, the undertaking concerned may prove an objective justification under Article 102 TFEU resp. fulfilment of the conditions of Article 101(3) TFEU, as applicable. Enforcement under both provisions is evidentially demanding and can take substantial time. Interim measures may sometimes preserve competition pending proceedings, but they are exceptional and do not provide a general framework for designing agentic market infrastructure. The DMA addresses part of that timing problem by imposing specified ex ante duties, but they only apply to designated gatekeepers.