Google appeals the EU's order to open Android to rival AI assistants
Google and Brussels are heading back to court, and this time the argument is about who gets to be the assistant on your phone.
Google filed an appeal at the EU's General Court on Tuesday against the Digital Markets Act decisions that would force it to share search data with rival search engines and open Android to competing AI assistants, arguing both mandates threaten user privacy and device security.
Oliver Bethell, Google's senior director for competition, said the decisions "will force us to share people's private search history without sufficient anonymisation and weaken vital security protections on Android," and warned of "irreversible harm to user privacy." Brussels set the timetable in July: Google must begin sharing search data from January 2027 to "rebalance the playing field" for rival search engines, while the Android changes are due next year and would let a competing assistant be activated by voice and asked to carry out tasks in other apps — the same reach Google's own Gemini already has. Google's answer is that Android is already open to other services, and that the search-data requirements lack adequate safeguards, user knowledge and consent. The Commission counters that data would be anonymised before it is shared, and that Google can assess whether a recipient poses serious security or data-protection risks. It is a separate proceeding from the €890 million fine Brussels handed Google in July for favouring its own services in search results.
The interesting part is which half of this fight bites in AI. Search data is an advertising and model-training story; assistant interoperability is a distribution story. If the Android order survives appeal, the default assistant stops being a product decision Google controls and becomes something rival labs can bid for at the operating-system layer — which is the only place a challenger can meet Gemini on reach rather than on benchmark scores. Google's framing is telling too: it is not arguing the changes are bad for competition, it is arguing they are unsafe, and it wants the court to treat privacy and security as the standard the Commission should have applied.
Manus shipped version 2.0 on Monday, turning its agent into a platform and launching Cue, a companion app that gives every personal agent its own email address, phone number, wallet and cloud computer.
The release is built around Cascade, Manus's in-house agent harness, which the company says used 23.2 percent fewer tokens, finished tasks 28.2 percent faster and cost 32 percent less in one tested configuration. Manus did not describe the test and no one outside the company has verified those figures, so treat them as a claim. The substance is elsewhere: Manus Studio adds editing workspaces for video and games, scheduled tasks become Automations that fire on incoming email, calendar, Slack or Notion events, and Computer Use lets an approved agent work with files and apps on the user's own machine, driven remotely from a phone. Cue agents can send messages, take calls and pay inside a budget the user sets, with several agents able to share one group chat and split a job.
This is the company's first substantial launch since Beijing blocked Meta's $2 billion acquisition and Manus went back to operating on its own — we covered that unwinding on the way out — Manus seeks $4B valuation in first round since Meta unwind. The caveat worth carrying into the launch week: Salt Labs disclosed in September that an email carrying an obfuscated instruction could run code inside a victim's Manus environment and reach credentials for connected apps like Gmail, Dropbox and GitHub. The flaw was triaged and patched through Meta's bug-bounty program. Agents with their own wallets and phone numbers widen exactly that attack surface, and the industry has not solved prompt injection yet — it has only learned to patch faster.
Mistral chief executive Arthur Mensch says the American safety debate has become cover for his competitors' negligence. "The debate that we've seen in the U.S. has been a cover for the negligence of some of our competitors," he told CNBC, arguing the actual requirement is monitoring and containment for agents that "go and do things that you do not expect." Mistral, unlike its US rivals, has no plans to slow development; Mensch says its next-generation model will close the gap with the leading labs "very significantly," backed by the €3 billion round Samsung led earlier this month.
He is half right, and the half that is self-serving is easy to spot. Agent containment is a genuine engineering problem rather than a marketing one, and his diagnosis lands the same week OpenAI shelved GPT-6.1 Astra because testing found the model going beyond authorized scope — an incident that reads less like caution theater and more like a lab discovering its sandbox is the product. A European challenger with a smaller compute budget has an obvious interest in arguing the safety question is a distraction, but then, so does every lab that would rather not answer it.
What to watch: whether the January 2027 search-data deadline holds while the appeal runs, and what Cue costs once its free early access ends.
Which would you rather decide your phone's default assistant — a court in Luxembourg, or the regulator that wrote the order? Tell us in the comments.
Sources: Euronews · Bloomberg via Techmeme · Malay Mail · Manus · The Decoder · Implicator.ai · CNBC · Forbes
