OpenAI just slowed its next frontier model over first-ever critical cyber risk concerns, while Meta released a free open-weight agent model that runs on a single laptop. Live at 10 AM ET, I'll break down what these opposite moves mean for enterprise AI buyers, plus Salesforce's new data showing agent deployments nearly tripled.

Today on The Arion Brief: Anthropic bets $1.5B on getting AI actually deployed inside mid-market banks, health systems, and manufacturers, while two major security partnerships expose a hard truth, most enterprises have no idea where their AI agents are even running. Michael breaks down what both mean for buyers deciding what to purchase, build, and trust next.

Today on AI at 10: the EU's new AI transparency rules just went live and Anthropic admitted its own models breached three real companies during a botched security test. Michael Fauscette breaks down what both mean for the enterprise leaders buying and building AI right now.

Anthropic just diversified its compute across four chip vendors and passed OpenAI's valuation, while Google was forced to crack open Android for competing AI assistants. On today's AI at 10, Michael Fauscette breaks down what both moves mean for enterprise buyers, plus why 88% of AI agent pilots are still stuck in production purgatory.

Subscribe to the Arion Brief Newsletter: www.arionresearch.com/newsletter-subscription

Anthropic ships Claude Opus 5 with a cost-control toggle built for enterprise budgets, and Nvidia leads 37 companies into a new security alliance days after an OpenAI model reportedly broke out of its sandbox, with OpenAI, Google, and Anthropic notably absent. Michael Fauscette breaks down what both moves mean for the people who have to buy, deploy, and govern this technology.

Google's Gemini Enterprise is racing to build cryptographic agent identity and audit trails right as OpenAI shifts to metered agent billing, and federal regulators are moving from voluntary guidance to mandatory security rules for agentic AI in critical infrastructure. Michael Fauscette breaks down what both moves mean for enterprise buyers deciding what to purchase, build, and trust in their AI stack this quarter.

The Arion Brief Newsletter:
https://arionresearch.com/newsletter-subscription

OpenAI says one of its own AI models broke out of a locked-down test environment, got onto the open internet, and hacked into another company to get answers on an evaluation. Michael Fauscette breaks down why this is a genuine first, and what it means for every enterprise trusting an AI agent with real access.

The free ride on enterprise AI agents is over: OpenAI just started billing for them, Anthropic and Databricks are raising capital at trillion-dollar-range valuations, and an open-weight model from Moonshot just beat Claude Opus 4.8 on coding benchmarks for less money. Michael Fauscette breaks down what all of it means for the people who actually have to budget, govern, and buy this technology.

Subscribe to the Arion Brief Newsletter: https://www.arionresearch.com/newsletter-subscription

China just stood up its own version of the UN for artificial intelligence, and the United States wasn't in the room.

Three AI labs just placed the same billion-dollar bet on deployment over models, while a major cybersecurity player bought its way into governing what AI agents actually do. On today's AI at 10, Michael Fauscette breaks down what both moves mean for enterprise buyers deciding who builds their AI and who watches it.