Just after regulators across Europe and the United States raised alarms, Anthropic is moving ahead with plans to give U.K. financial institutions controlled access to its most advanced cybersecurity model, Mythos.

The rollout, expected within days, marks a new phase of Project Glasswing, the company’s tightly controlled initiative designed to equip defenders with frontier AI tools before such capabilities become widely available.

A Rapid Escalation in AI Security

Mythos has drawn intense attention for its ability to identify and chain together zero-day vulnerabilities across major operating systems and web browsers.

That capability has triggered urgent responses from regulators. In the U.S., officials including Treasury leadership and the Federal Reserve have already held high-level meetings with major banks. In Europe, authorities such as the Bank of England, Financial Conduct Authority, and national cybersecurity agencies have been actively assessing the risks.

The concern is straightforward. A tool that can systematically uncover hidden weaknesses in critical infrastructure could dramatically improve defenses, but it could also accelerate attacks if misused.

Controlled Access, Not Public Release

Anthropic has made clear that Mythos will not be broadly released. Instead, access is being tightly restricted to vetted organizations through Project Glasswing.

Participants already include major players such as JPMorgan Chase, Amazon Web Services, Google, and Microsoft.

The addition of U.K. banks expands that circle, bringing one of the world’s most interconnected financial systems into direct engagement with frontier AI security tools.

A Double-Edged Tool for Banks

For financial institutions, Mythos presents both an opportunity and a threat.

Used defensively, it could function as a powerful red-team system, uncovering vulnerabilities before attackers exploit them. That could significantly shorten response times and strengthen resilience across complex banking infrastructure.

At the same time, the very capabilities that make Mythos valuable also increase the stakes. If similar tools become accessible to malicious actors, the scale and speed of cyberattacks could rise sharply.

Industry Braces for an AI-Native Era

Anthropic CEO Dario Amodei has described the current moment as a transitional period where risks are real but manageable with the right safeguards. The company has emphasized that the threat is no longer theoretical.

According to Anthropic’s regional leadership, engagement from U.K. bank executives has intensified in recent days, reflecting growing awareness that AI-driven cybersecurity is no longer a future concern.

The Bigger Shift

The rollout of Mythos into the banking sector signals a broader turning point.

AI is no longer just a productivity tool or research breakthrough. It is becoming a core component of national and financial security infrastructure.

For banks in the U.K. and beyond, the message is increasingly clear. The era of AI-native cybersecurity has begun, and adaptation is no longer optional.

AI & Machine Learning, Cybersecurity & Privacy, News

The European Commission is stepping up pressure on Google, proposing new rules that would require the company to share portions of its search data with competitors, including AI-powered services.

The move is part of ongoing enforcement of the Digital Markets Act, which aims to curb the dominance of large technology platforms and open up competition across digital markets.

Opening Google’s Data to Rivals

Under the proposal, Google would be required to give third-party search engines and AI tools access to certain search data. This includes defining how frequently data must be shared, how it is anonymized, and how access is priced.

The goal is to allow smaller competitors to improve their own search services and better compete with Google Search, which remains the dominant gateway to information online.

Importantly, the rules could also apply to AI chatbots with search capabilities, a fast-growing category as companies race to integrate real-time information into conversational interfaces.

Privacy Concerns Take Center Stage

Google has strongly pushed back against the proposal, arguing that sharing search data could put user privacy at risk.

The company says Europeans rely on its platform for sensitive queries related to health, finances, and personal matters, and that handing data to third parties could weaken protections despite anonymization requirements.

This tension highlights a core challenge for regulators. Increasing competition often requires data sharing, but search data is among the most sensitive categories of user information.

A High-Stakes Regulatory Battle

The European Commission is now consulting stakeholders, with feedback open until May 1 and a final decision expected in July.

The proposal follows earlier charges against Google for alleged violations of the Digital Markets Act. While the company has offered concessions, rivals have argued those measures do not go far enough.

Financial stakes are significant. Under the DMA, companies can face fines of up to 10 percent of global annual revenue for non-compliance. Google has already accumulated nearly €10 billion in antitrust fines in Europe since 2017.

AI Changes the Equation

The inclusion of AI-powered search tools adds a new dimension to the dispute.

As AI chatbots increasingly act as intermediaries for information retrieval, access to high-quality search data becomes a competitive advantage. Regulators appear to be anticipating this shift, aiming to ensure that emerging AI players are not locked out by incumbents controlling critical data.

The outcome of the case could shape not just traditional search competition, but also how data is shared in the next generation of AI-driven services.

AI & Machine Learning, News, Regulation & Policy

Anthropic is expanding its footprint in London, securing new office space that can accommodate up to 800 employees as competition for AI talent intensifies across Europe.

The company currently has more than 200 staff in the city, but the new expansion signals a long-term commitment to building one of its key hubs outside the United States.

A Growing AI Hub in London

Anthropic’s new office will be located in London’s Knowledge Quarter, a fast-growing cluster of AI research, startups, and academic institutions. The area already hosts major players including OpenAI, Google DeepMind, and Meta, along with companies like Synthesia and Wayve.

The move comes just days after OpenAI announced plans for its first permanent office in London, underscoring how central the U.K. has become in the global AI race.

Anthropic’s Pip White said the expansion reflects both the strength of local talent and the U.K.’s focus on AI safety and regulation, two areas increasingly shaping where companies choose to invest.

Momentum Behind the Business

The expansion follows a period of rapid growth for Anthropic. The company’s annualized revenue has surpassed $30 billion, with more than 1,000 enterprise customers each spending over $1 million annually.

Its valuation has surged as well, reaching $380 billion earlier this year, with reports suggesting investor interest at even higher levels.

Recent product launches have helped fuel that momentum, including its coding-focused tools and advanced models designed for tasks like software analysis and security.

Strategy and Geopolitics

The London expansion also comes amid shifting geopolitical dynamics. U.K. officials have reportedly been courting Anthropic more aggressively following its high-profile dispute with the U.S. Department of Defense over how its AI models could be used.

Establishing a larger presence in London gives Anthropic access not only to talent, but also to a regulatory environment that is actively shaping its approach to AI governance and safety.

Talent Is the New Battleground

As AI development becomes more compute-intensive and capital-heavy, talent remains one of the few truly scarce resources.

Cities like London are emerging as key battlegrounds, where top researchers, engineers, and product teams are increasingly concentrated. Companies are responding by expanding physical hubs, even as much of AI work can technically be done remotely.

Anthropic’s move makes clear that the race is not just about models and infrastructure, but also about where the people building them choose to work.

AI & Machine Learning, News

Strong outlooks from ASML and TSMC are reinforcing a clear message across the industry. The AI infrastructure boom is far from slowing down.

Both companies raised their forecasts this week, pointing to sustained demand for advanced chips used in training and running large AI models. The signals suggest that major cloud providers are continuing to invest heavily in compute, even as questions grow about returns on those investments.

AI Spending Machine Keeps Running

Executives say demand is still being driven by hyperscale customers. Companies like Microsoft, Amazon, and Meta are expected to collectively spend more than $600 billion this year on data centers and AI infrastructure.

TSMC CEO C.C. Wei pointed to strong signals across the supply chain, noting that demand is not only coming from direct customers but also from their downstream clients. In practice, that means cloud providers racing to secure chips before competitors do.

This demand flows directly to chip designers such as Nvidia, AMD, and Broadcom, all of which rely heavily on TSMC’s manufacturing capacity.

The Bottleneck Problem

Despite strong demand, the industry faces a fundamental constraint. There are only a few companies capable of producing cutting-edge chips at scale.

ASML, which supplies the lithography machines required to manufacture advanced semiconductors, expects demand to exceed supply for the foreseeable future. That constraint is now affecting not just AI, but also smartphones and PCs.

TSMC echoed similar concerns, noting that capacity remains tight even as the company ramps up capital spending to expand production.

This dynamic has pushed companies toward long-term agreements to lock in manufacturing capacity, sometimes years in advance. Securing supply has become as critical as designing the chips themselves.

Shift Toward Advanced AI Chips

Another notable shift is where demand is concentrating. Increasingly, spending is moving toward high-performance processors used for inference, the stage where trained AI models generate real-world outputs.

This reflects the next phase of AI adoption. After an initial wave focused on training large models, companies are now scaling deployment across products and services, which requires massive inference capacity.

Boom With Questions Attached

Even as forecasts remain strong, investor pressure is building. Markets are increasingly focused on whether massive AI spending will translate into meaningful returns.

Some analysts warn that the current pace of investment could eventually face limits, especially if monetization lags behind infrastructure build-out.

For now, however, the signals from ASML and TSMC suggest the opposite. Demand is still accelerating, supply is constrained, and the race for AI compute is intensifying across the entire technology stack.

AI & Machine Learning, Cloud & Infrastructure, News