Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
The European Commission has specified how Google must open two assets that reinforce its position in AI: privileged Android capabilities and search feedback gathered at scale. The decisions create enforceable routes for competitors, but they do not give rivals immediate access or establish that the resulting products will be useful, affordable or widely adopted.
The Commission adopted the final measures on July 16 under the Digital Markets Act. One requires Google to provide competing AI services with free, effective access to covered Android features on terms comparable to those available to Google's services. The other requires it to offer eligible online search engines—including AI chatbots with online search functions—an anonymized set of Google Search data under fair, reasonable and nondiscriminatory terms.
The comparison matters. The Android order is about operating-system capabilities, not access to Gemini as a service. The Search order is about signals that can help a competitor improve query understanding, retrieval, ranking and indexing, not a copy of Google's search engine.
The Commission says around 60% of European mobile users have Android devices. It also says Google Search has held more than 90% of the European search market for decades and uses its volume of user data to improve search. Those are the Commission's stated bases for treating distribution and feedback data as barriers to competition; they do not show how much either order will narrow Google's lead.
The approach is not confined to Google. Brussels has also required Apple to add interoperability features connecting its devices with non-Apple products, as a contemporaneous account of the Google decisions notes. The new measures apply that regulatory method to mobile AI services and search data.
The Android decision covers 11 features in four functional layers:
| Layer | What third-party AI services are due to receive | Important limit |
|---|---|---|
| Invocation | Access through the home control and always-on wake words | Concurrent wake words have the latest deadline |
| Context | Central access to on-device app data chosen for sharing, proactive context and real-time microphone, camera, screen and speaker inputs | Users must consent; apps decide which stored data to expose |
| Actions | Structured app actions, multi-step screen automation and operating-system controls | Apps and users choose available tasks; sensitive features can require certification |
| Computing resources | Comparable use of system-level on-device models, conditions for running third-party on-device models and background execution | Parity applies to covered system resources, not to Gemini as a whole |
This reaches well beyond permission to install or launch another assistant. A rival could be invoked when the screen is off, act inside other apps in a separate virtual window, change device settings or use local models under conditions comparable to Google's services. Google must supply documentation, testing and technical assistance, and make later functionality in the covered features available to third parties when it becomes available to Google.
The order nevertheless leaves Google responsible for building the interfaces. It permits objective and nondiscriminatory eligibility conditions for five sensitive areas: screen automation, structured app integration, system integration, centralized access to on-device app data and context-aware intelligence. Independent parties are to certify qualifying apps alongside Google. Draft eligibility terms are due by February 1, 2027, final terms and applications by May 1, and assessments must be completed within four weeks after an application is received.
User consent applies to access across all 11 features. The measures also do not require phone makers to add hardware: interoperability applies where a device already has the necessary technical capacity, and Google cannot shift its implementation work to manufacturers.
Manufacturers remain free to customize Android, preinstall apps, set defaults during device setup and monetize that promotional space, provided they do not obstruct interoperability. Meanwhile, Google cannot make access to the 11 features depend on a rival assistant holding a default role. The order therefore targets technical parity without eliminating every commercial distribution advantage.
Most features must be implemented in Android 18 by August 1, 2027. Concurrent hotword detection, which would allow multiple assistants to listen for their respective wake words, is due in Android 19 by August 1, 2028. The Commission therefore has set a parity standard well before it will be possible to judge the shipped interfaces.
Google disputes both the safety of that standard and the premise that rivals lack a route into Android. Kent Walker, president of global affairs at Google and Alphabet, said in the company's response that AI assistants already access Android features through a process in which phone manufacturers help validate them. He argued that granting powerful permissions without those safeguards threatens device security and called for flexible, test-based implementation.
The Commission's position is narrower than a claim that access is risk-free. It says consent remains necessary, EU privacy and cybersecurity rules continue to apply, and certification can protect the most sensitive capabilities. Whether those controls deliver comparable access without introducing unacceptable risk remains an implementation question, not a settled fact.
The Search decision covers anonymized ranking, query, click and view data produced by free and paid search. Depending on the record, the shared fields can include query language and device type, viewed URLs, interactions with results and result position. Eligible recipients can use the dataset to develop and optimize online search services.
They cannot use it to train general-purpose AI models, build consumer-profiling or advertising services, or systematically reproduce Google's results. The order does not require Google to disclose its algorithms or technology. Nor does it provide user-account information or search histories; the shared material is a more restricted and heavily altered subset of the data Google collects.
That distinction both limits the competitive value and frames the privacy dispute. Alphabet must remove direct identifiers such as usernames and IP addresses, strip selected metadata, suppress unusually long queries and records containing rare terms, and generalize location, device type and query language. Every end user must fall within a group of at least 1,000 people sharing those generalized attributes, while 95% must fall within groups of at least 29,000.
Contracts must bar recipients from linking the data to other datasets, disclosing it onward, attempting re-identification or reversing the technical protections. Applicants need an independent audit before access, another within six months after processing begins, and annual audits thereafter.
Those measures do not make the entire dataset legally or practically innocuous. The Commission says queries can still contain personal data about the people being searched for, even after the person who issued a query has been anonymized, so recipients remain subject to the GDPR for that material. The framework is also subject to reviews every two years and can be reopened if material facts change, including new evidence about re-identification attacks.
Google's objection is more categorical. Walker said private searches could be exposed to unfamiliar companies without adequate anonymization or users' knowledge and consent, risking individual privacy, business secrets and national security. The available evidence establishes the safeguards the Commission ordered and Google's challenge to their adequacy; it does not establish how they will perform against real recipients and attacks.
This is not an open or free public dataset. Applicants must operate an online search service, have averaged at least 50,000 monthly EU users during the previous year, and satisfy security, sanctions and international-transfer conditions. An established applicant must have provided search in the EU for at least two consecutive years. A company founded less than two years earlier can instead qualify as a credible new entrant only if it has received more than €50 million in capital investment.
That financing threshold does not apply to every recipient, but it means a newly formed search company needs substantial backing before it can qualify. The scale, audit and secure-processing requirements further concentrate practical access among companies able to fund compliance and search infrastructure.
Alphabet may charge each beneficiary. The baseline permits recovery of incremental preparation, storage, transmission and onboarding costs, plus a reasonable return on the capital needed to provide the data, capped at Alphabet's weighted average cost of capital. Charges include fixed and variable components; one-time common costs will be divided using the number of applicants that begin an audit or are found eligible within six months of adoption, while recurring common costs are divided among recipients each year.
Contrary to a simple cost-recovery description, exceptional circumstances can support an additional margin, limited by Google Search's operating margin. The Commission says that exception could apply if Alphabet's commercial use cannot cover efficiently incurred collection costs or if a recipient itself operates at gatekeeper scale; it cannot be applied to micro, small or medium-sized enterprises.
The data will also arrive with at least a seven-day delay, and each beneficiary can receive it for no more than five years. Those terms are designed to push recipients to develop their own search technology rather than operate a live replica, but they also make the dataset a bounded input rather than an enduring substitute for collecting first-party search behavior.
The stricter framework follows a failed first attempt. The Commission says Alphabet's initial proposal suppressed 90% to 100% of unique queries, excluded AI chatbots that offer search and produced no meaningful uptake. That history is the clearest warning against equating a formal access right with an effective one.
The first measurable test will arrive on Search. Alphabet must publish application information by the end of August 2026, provide template licenses and test samples by September, finalize the anonymized dataset by November, and communicate a final pricing offer by January 2027. The samples will include a small real-data set, a synthetic set and a larger representative set; only prospective recipients with an auditor's assurance report can access the larger one. The decisive evidence will be how many companies pass eligibility and auditing, what those samples reveal about utility, what Google charges and whether any recipient improves a search product with the dataset.
Android will take longer. Developers will need to test whether Google's interfaces match the access available to its own services without extra friction, whether certification is timely and nondiscriminatory, and whether user consent works clearly for capabilities as sensitive as ambient sensors and cross-app actions. The Commission says it will monitor Google's design, development and release work over the next two years.
Those results will resolve the central question the decisions themselves cannot: whether regulated access can loosen Google's control of mobile AI distribution and search feedback without making the data too degraded, the implementation too risky or the economics too demanding for rivals to use it.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.
Runta has raised a $20 million seed round led by Andreessen Horowitz for an execution layer that governs agents at the operating-system and network levels, entering a market where cloud runtimes and sandbox providers already offer overlapping controls.