Anthropic leads the enterprise market at HumanX

At this week's HumanX conference in San Francisco, Anthropic — not OpenAI — was the name on everyone's lips. The company now leads OpenAI in enterprise spending across finance, insurance, and information services, and Claude Code alone is generating over $2.5 billion in annualized revenue. Anthropic also announced Claude Managed Agents, a new framework for deploying persistent AI agents within enterprise workflows — a direct play for the agentic enterprise market that consultancies should be tracking closely.

For business leaders evaluating AI vendors, this signals a genuine three-way race emerging. Your enterprise AI strategy can no longer assume OpenAI dominance — the competitive landscape has shifted measurably, and the winners will be the organizations that treat vendor diversity as a strategic asset rather than a compliance headache.

Anthropic's "Mythos" model triggers emergency Wall Street meeting

Treasury Secretary Bessent and Fed Chair Powell convened an emergency session with major bank CEOs after Anthropic's frontier model preview — internally called "Mythos" — discovered exploitable vulnerabilities in every major operating system and browser during testing. Anthropic has committed $100 million in credits to help defenders patch critical systems first.

This is a landmark moment for the dual-use nature of frontier AI: the same capabilities that find vulnerabilities can also be weaponized, and it's forcing policymakers and financial institutions to rethink their cyber resilience frameworks in real time. For enterprise security leaders, this underscores a hard truth — frontier AI has arrived at the point where it poses genuine systemic risks, and your organization's AI governance framework needs to account for vulnerabilities, not just productivity gains.

Meta goes proprietary with Muse Spark

Meta's first major model under chief AI officer Alexandr Wang, Muse Spark, has catapulted the Meta AI app to No. 2 on the App Store — surpassing ChatGPT. But the bigger story for enterprise leaders is what's missing: unlike every previous Llama model, Muse Spark is entirely proprietary. Meta has effectively exited the open-weight frontier race, a strategic shift that reshapes the competitive dynamics of enterprise AI procurement and model choice.

This move signals something important: the economics of frontier model development are pushing even champions of open-source toward a closed model. If you've been betting on open-weight alternatives for vendor independence, you may need to recalibrate that strategy.

EY deploys enterprise-scale agentic AI into the audit

EY has rolled out a multi-agent AI framework embedded directly in its Canvas audit platform, built on Microsoft Foundry, Fabric, and Azure. The firm is among the inaugural 14 organizations in the Frontier Firm AI Initiative deploying advanced AI at scale. For the professional services world, this is a signal that the competitive bar for audit and assurance is being fundamentally reset — firms that aren't building agentic capabilities into core service delivery risk falling behind fast.

Q1 venture funding hits $300 billion — 80% flows to AI

Global venture investment reached an all-time high of $300 billion in Q1 2026, with $242 billion — a staggering 80% — flowing to AI companies. The four largest rounds went to OpenAI ($122B), Anthropic ($30B), xAI ($20B), and Waymo ($16B), collectively accounting for 65% of all global venture dollars. The capital concentration at the frontier raises real questions about market structure, competitive moats, and what this means for the broader startup ecosystem outside of AI.

Google releases Gemma 4 for agentic workflows

Google released Gemma 4 under the Apache 2.0 license — its latest series of open models built specifically for advanced reasoning and agentic workflows. With Meta going proprietary, Google is now arguably the most important player keeping the open-weight frontier alive. For enterprise teams building custom agent pipelines, Gemma 4 represents a strong new option that avoids vendor lock-in.

Gemini 3 Deep Think achieves gold-medal science reasoning

Google upgraded Gemini 3 Deep Think with breakthrough scientific reasoning capabilities, achieving gold-medal performance on the 2025 International Physics and Chemistry Olympiads. Beyond academic benchmarks, the practical applications are noteworthy: users can now sketch a concept and have Deep Think generate a 3D-printable file. This signals the accelerating convergence of reasoning models with real-world engineering and design workflows.

California asserts state-level AI regulatory authority

After the Department of Defense flagged Anthropic as a supply-chain risk, California Governor Newsom announced the state will conduct its own independent review and make its own business decisions regarding AI companies. Meanwhile, Washington state passed two major AI bills on disclosure requirements and chatbot safety. The regulatory landscape is fragmenting rapidly, and enterprises operating across jurisdictions will need to build governance frameworks that can adapt to overlapping — and sometimes conflicting — requirements.

One Thing to Think About

This week crystallized something that's been building for months: the agentic AI era isn't arriving — it's here, and the early winners are the organizations treating agents as core infrastructure rather than experimental toys. EY embedding multi-agent systems directly into its audit platform, Anthropic launching Managed Agents, OpenAI's Frontier helping Fortune 500 companies deploy agents company-wide — these aren't pilot programs anymore.

The real competitive question for 2026 isn't "should we adopt agentic AI?" but "what parts of our value chain do we redesign around agents first?" The organizations that answer that question well in the next 12 months will define industry benchmarks for the next decade.

Worth Your Time

Nathan Lambert: The Inevitable Need for an Open Model Consortium — AI researcher Nathan Lambert argues that the economics of frontier model training are pushing the industry toward an inevitable consortium model for open AI. With Meta going proprietary, Chinese open-model startups facing financial pressure, and training costs hitting the billions, Lambert predicts companies reliant on open-weight models will eventually pool resources — much like industry consortia in semiconductor R&D. A must-read for anyone advising clients on long-term AI sourcing strategy.

Enterprise Agentic AI Landscape 2026: Trust, Flexibility, and Vendor Lock-in — Kai Waehner's comprehensive analysis of the agentic AI vendor landscape, covering trust frameworks, multi-model strategies, and how enterprises can avoid getting locked in as the market consolidates.

From Assistant to Actor: What the Rise of Agentic AI Means for Your Business — Morgan Lewis's legal take on the contractual, liability, and governance implications of deploying AI agents that act autonomously on behalf of your organization. Essential reading for anyone advising clients on agentic deployment risks.

OpenAI Enterprise Revenue Hits 40% — Parity with Consumer by Year-End — Enterprise now makes up more than 40% of OpenAI's revenue and is on track to reach parity with consumer by the end of 2026. OpenAI Frontier is helping customers like Oracle, State Farm, and Uber build and manage company-wide agent deployments. GPT-5.4, with its 1-million-token context window and ability to autonomously execute multi-step workflows, scored 75% on the OSWorld-V desktop productivity benchmark — a meaningful step toward agents that can genuinely work across enterprise software environments.

That's your briefing for Saturday. See you Monday morning.

— Chiel