Regulation Tightens as Anthropic Eyes a Trillion-Dollar IPO

Two sweeping European mandates, a landmark AI IPO filing, a high-profile model delay, a new White House cybersecurity initiative, and a legal crackdown on deepfake apps defined the AI news cycle this week.

EU Orders Google to Open Android AI to Rivals

The European Commission delivered one of its most consequential rulings under the Digital Markets Act this week, ordering Google to grant competing AI assistants direct access to Android's camera, microphone, and on-screen content by August 1, 2027. The ruling also requires Google to share anonymised search query data with rival search engines — a measure that could materially shift the competitive landscape for AI-powered search.

The order covers Android 18 and is expected to affect roughly 60% of EU Android users. French startup Mistral, among other European AI companies, is widely seen as a primary beneficiary. Google has until the end of this month to submit a compliance roadmap.

The decision is the most direct intervention yet into Google's control of the Android ecosystem, forcing the company to treat its own AI assistant no differently than third-party alternatives. Critics of the DMA had long argued that Android's tight hardware integration gave Google AI an insurmountable head start; the Commission has now agreed.

EU AI Act: Transparency Rules Take Effect August 2

While the full weight of the EU AI Act will not land until 2027–2028, the first significant enforcement wave begins on August 2, 2026. Article 50 of the Act imposes new transparency obligations: chatbots must disclose that users are interacting with an AI, AI-generated content must be labelled, and deepfake imagery must carry explicit markers.

Simultaneously, the European Commission gains formal authority to enforce rules governing General-Purpose AI models — the category that covers large language models like GPT and Claude. Non-compliance can trigger fines of up to €15 million or 3% of global annual turnover, whichever is higher.

Despite a parliamentary delay to the high-risk AI deadlines, the August 2 provisions are firm. Companies that have been waiting on a unified EU deadline are now running out of runway. Compliance teams across the continent have been warned: internal audits and vendor questionnaires will not be enough — live product changes must be in place before the deadline.

Anthropic Files Confidentially for IPO

Anthropic has submitted a confidential draft registration statement to the U.S. Securities and Exchange Commission, signalling that a public offering could arrive before year's end. According to reports, the company is tracking roughly $47 billion in annualised revenue and has recently crossed into profitability — a milestone few AI labs have reached at this scale.

The company is also reportedly arranging several billion dollars in additional credit facilities ahead of the listing. Analysts have speculated that a market debut could value Anthropic near $1 trillion, which would rank it among the largest technology IPOs on record, comparable in scale to the listings of Meta and Alibaba.

Anthropic's filing comes at a moment of intense investor interest in AI infrastructure. The company's Claude model family has gained significant enterprise traction, and its public focus on AI safety has resonated with institutional buyers wary of reputational risk. The confidential S-1 process allows the company to negotiate terms and disclose financials on its own timeline.

Google Delays Gemini 3.5 Pro After Internal Failures

Google has postponed the release of Gemini 3.5 Pro after the model reportedly fell short on coding and complex reasoning benchmarks in internal testing. Sources familiar with the matter indicate that engineers discovered structural failures in the model's handling of recursive tool-calling — a critical capability for agentic AI applications.

Rather than ship a patched version, Google is said to have scrapped the original base model and restarted pretraining from scratch. The news rattled investors: Alphabet shares fell roughly 4%, erasing tens of billions in market capitalisation.

The stumble is notable given the competitive pressure Google faces from OpenAI and Anthropic. Gemini 3.5 Pro was widely expected to help Google close the gap on complex reasoning tasks. A revised release timeline has not been announced.

White House Launches GOLD EAGLE Cybersecurity Clearinghouse

On July 14, the White House unveiled GOLD EAGLE, an AI-powered clearinghouse designed to accelerate the detection and remediation of software vulnerabilities across critical infrastructure. The initiative implements President Trump's June 2 Executive Order on cybersecurity and brings together the Treasury Department, the Department of Homeland Security's CISA division, and the Department of Defense alongside private-sector critical-infrastructure operators.

The programme aims to shrink the gap between vulnerability discovery and coordinated disclosure — a window that adversaries have historically exploited. By using AI to triage and correlate vulnerability reports across sectors, GOLD EAGLE is designed to give defenders a faster response timeline.

Financial institutions are paying particular attention, with briefings already underway on how the clearinghouse will interact with existing regulatory reporting obligations. The initiative represents one of the most direct deployments of AI at the federal level for defensive cybersecurity purposes.

San Francisco Moves Against AI Nudify Apps

San Francisco City Attorney David Chiu sent cease-and-desist letters to both Apple and Google this week, demanding removal of 13 AI applications used to create non-consensual intimate deepfakes — commonly referred to as "nudify" apps. Chiu argued that both platforms profit from distributing tools used primarily to exploit women and girls.

The platforms responded swiftly. Google suspended five apps from the Play Store and reviewed hundreds of related listings. Apple removed three apps and began terminating the developer accounts behind them. The legal letters stopped short of formal litigation but established a clear escalation path if the companies fail to act comprehensively.

The action fits within a growing wave of state and local enforcement targeting AI-generated non-consensual intimate imagery. Several U.S. states have already enacted criminal statutes in this area, and the San Francisco move signals that platform liability is increasingly on the table. Notably, the EU's AI Act deepfake labelling requirements — taking effect August 2 — directly cover this category of application.