AI Daily 11 Apr: Meta Launches SI Labs, Anthropic Hits $30B

The big AI stories this week were not about controversy. They were about capital, capability and scale. Meta debuted the first model from a brand new research division. Anthropic crossed $30 billion in annualised revenue. And Google DeepMind formally embedded its AI tools inside every major US government science facility. If the question in 2024 was whether generative AI would find real commercial traction, the answer in April 2026 is firmly yes, and the pace is accelerating.

Meta has launched Muse Spark, the first model from its new Meta Superintelligence Labs division. The division was formed following Meta’s $14 billion arrangement to bring in Scale AI chief Alexandr Wang, signalling a deliberate push into the frontier research space previously dominated by Anthropic, OpenAI and Google DeepMind. Muse Spark is described as achieving comparable performance to Llama 4 Maverick at a fraction of the compute cost, more than an order of magnitude less, according to Meta’s own framing. For developers and businesses building AI applications, that kind of efficiency improvement has direct implications for cost and accessibility.

The model was unveiled at LlamaCon, Meta’s first developer conference. Alongside Muse Spark, Meta previewed the Llama API, a platform designed to make it straightforward to build Llama-based applications, with one click API key creation and interactive model playgrounds. For UK developers who have been building on open-weight Llama models, both announcements are relevant. Meta AI is already available in the UK across WhatsApp, Messenger and Instagram, and the Muse series signals that Meta’s open ecosystem intends to stay competitive with proprietary lab offerings. Source: Meta AI Blog.

Anthropic’s annualised revenue has surpassed $30 billion, up from roughly $9 billion at the end of 2025. More than 1,000 business customers are now spending over $1 million per year on Claude, a figure that reportedly doubled in under two months. Alongside this, Anthropic announced an expansion of its compute partnership with Google and Broadcom, covering multiple gigawatts of next-generation processing capacity, and separately committed $50 billion to American AI infrastructure.

For UK businesses currently evaluating AI vendors, numbers at this scale matter. Anthropic is no longer a research project competing on model quality alone. The commercial infrastructure covers enterprise contracts, data residency commitments and compliance positioning that matches what the largest cloud providers offer. The compute expansion also signals that Anthropic is building towards running significantly more capable or more numerous models within the coming year. Source: Anthropic.

OpenAI’s enterprise segment now accounts for more than 40 percent of its total revenue, with parity with consumer revenue expected by the end of 2026. OpenAI published what it is calling its next phase of enterprise AI this week, covering expanded security features, deeper integrations for large organisations and new data handling commitments. The company also closed a $122 billion funding round in April, the largest private raise in technology history.

The enterprise shift matters because business customers fund very different product priorities than consumers do. They require auditability, service level guarantees, on-premise options and compliance controls. The security update published alongside the revenue milestone addresses those requirements directly. For UK IT teams evaluating ChatGPT Enterprise contracts, or businesses accessing OpenAI models via Microsoft 365 Copilot licences, the expanded security commitments are worth reviewing in full. Source: OpenAI.

Google DeepMind has formally deployed its AI tools across all 17 US Department of Energy National Laboratories. The deployment is part of the White House Genesis Mission, a programme to use AI to accelerate scientific research across energy, medicine and national security. Scientists at every DOE lab now have access to AI co-scientist, a multi-agent research tool built on Gemini designed to synthesise academic literature and generate novel hypotheses. The programme will expand to include AlphaEvolve, a Gemini-powered algorithm design agent, and AlphaGenome, a model for non-coding DNA analysis.

The scale of this deployment is significant. Embedding a commercial AI system into all 17 national laboratories is the largest formal integration of its kind in US federal scientific infrastructure. Tools that prove useful at this scale tend to find their way into commercial products. AlphaEvolve’s applications in materials science and energy research are particularly relevant to industries tracking advanced manufacturing and clean energy. The UK government has equivalent commitments through DSIT and the Alan Turing Institute. Expect pressure on both to formalise comparable arrangements with UK research institutions. Source: Google DeepMind.

UK financial regulators have confirmed they will not introduce AI-specific rules in 2026. The Financial Conduct Authority, Prudential Regulation Authority and Bank of England are maintaining a technology-neutral approach, applying existing conduct and prudential rules to AI contexts rather than creating a new framework. The FCA has launched the Mills Review, an ongoing assessment of how AI is reshaping retail financial services. Three active support mechanisms are now running alongside it: a Supercharged Sandbox offering computational resources for AI testing, AI Live Testing for controlled real-world trials, and an AI Sprint programme.

A formal UK AI Bill is not expected until after the next King’s Speech, at the earliest in the second half of 2026. For UK financial services firms, this means operating under existing rules with no new AI-specific compliance burden this year. The FCA’s sandbox and testing programmes are live now and open to firms wanting to test AI-driven products under regulatory oversight. Source: Results Sense / GOV.UK.

Here is everything else worth knowing from today’s AI news.

• Anthropic Claude Managed Agents in public beta. Anthropic launched a fully managed agent harness for running Claude as an autonomous agent with secure sandboxing and built-in tools. Aimed at developers building agentic applications via the API. Source: Anthropic Docs.
• Anthropic ant CLI launched. A command-line client for the Claude API enabling faster interaction and native integration with Claude Code, with YAML-based API resource versioning. Source: Anthropic Docs.
• Claude Haiku 3 deprecation set for 19 April 2026. Developers using Haiku 3 via the API should migrate before the retirement date. Source: Anthropic.
• OpenAI Child Safety Blueprint and Safety Fellowship published. OpenAI published its Child Safety Blueprint and announced an OpenAI Safety Fellowship on 8 April 2026. Source: OpenAI.
• Google NotebookLM integrated into Gemini. Google has embedded NotebookLM, its AI research tool, directly into the Gemini interface. Users can now upload PDFs, YouTube videos and web URLs through Gemini’s side panel. Source: Google DeepMind.
• Meta LlamaCon hackathon results and Llama startup programme details published. Meta shared the first LlamaCon hackathon winners and announced the Llama Startup Programme, awarding over $1.5 million to international recipients. Source: Meta AI.
• UK deepfake legislation in force from 6 February 2026. Amendments to the Sexual Offences Act 2003 now criminalise the creation of sexually explicit deepfake images of adults without consent. Source: UK Government.
• Anthropic Claude Mythos Preview remains gated. The cybersecurity-focused Mythos model is still invitation-only for defensive security researchers, available through Project Glasswing. Source: red.anthropic.com.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every day.