2026-09-02
By Vadym · Generated with AI, curated by me
Today’s AI news moved out of the model-release cycle and into institutions — the Pentagon rolled ChatGPT and Grok out to three million staff while pointedly excluding Anthropic, and the EU handed ChatGPT the same regulatory scrutiny as a search engine. Underneath the politics, the money and the plumbing keep moving: another $35 billion in Anthropic compute, a CrowdStrike partner program, and a robotics startup framing houses instead of chatbots.
On August 31 the Pentagon added OpenAI’s ChatGPT Mil and xAI’s Grok for Government to GenAI.mil, its secure internal AI portal, opening access to all 3 million DoD military and civilian personnel — 1.7 million have already signed on since the platform launched last year. ChatGPT Mil targets routine unclassified work like admin tasks and logistics. The rollout notably excludes Anthropic’s Claude, after the Trump administration designated Anthropic a supply-chain risk over the company’s refusal to strip safety guardrails from its tools.
The European Commission classified ChatGPT as a Very Large Online Search Engine under the Digital Services Act — the first time an AI chatbot has been designated at this tier — after OpenAI’s tool passed the 45-million-monthly-user threshold with roughly 159 million users in the EU. Reddit (57.2 million users) and Roblox (about 48 million) were designated Very Large Online Platforms in the same ruling. All three now have until the end of November to complete systemic risk assessments, submit to independent audits, and share data with regulators.
Anthropic signed a six-year, $35 billion agreement with cloud provider Lambda for computing capacity out of a Texas data center under construction by Hut 8. Nvidia — an investor in both Anthropic and Lambda — will lease the facility from Hut 8 and make it available to Lambda. The deal adds to at least $135 billion in compute contracts Anthropic has signed this year, including a separate $45 billion deal with Nscale for capacity in West Virginia.
OpenAI released Codex CLI 0.152.0 on September 1, adding vim-mode / and ? search with highlighted matches inside drafts, rate-limit banners with direct actions for checking usage or managing credits, and visible credential-refresh progress for Amazon Bedrock reauthentication. MCP server names can now include colons, @, slashes, and periods, supporting package-style naming across CLI commands. A same-day patch, 0.152.1, fixed a Guardian approval bug around Node REPL policies.
CrowdStrike introduced an AI Partner Specialization inside its Accelerate Partner Program at Fal.Con 2026, giving partners four paths: resell AI security through Falcon Flex, manage deployments, build agents with Charlotte AI AgentWorks and Falcon Foundry, or deliver full agentic transformations. It also launched a Verified Agent certification for partner-built agents entering the CrowdStrike Marketplace, with Accenture, CoreWeave, and World Wide Technology among the first named partners.
Reframe Systems, founded by former Amazon Robotics engineers, raised $40 million led by Energy Impact Partners to expand its network of automated homebuilding microfactories — smaller regional plants that use software and robotics instead of one centralized facility. Its new Billerica, Massachusetts facility is designed to produce up to 500 multifamily units or 250 single-family homes a year, and the company claims it can deliver homes three times faster and 35% cheaper than traditional construction.
Today’s common thread is AI showing up in rooms it didn’t used to be in — the Pentagon’s procurement office, the EU’s regulatory wing, a homebuilder’s factory floor — and each of those rooms is starting to write its own rules for what AI gets to do there. The model layer is still moving fast (another $35 billion in compute says so), but the more interesting fights are happening one layer up, in who gets to deploy it and under what conditions.
— Boba
Curated by Vadym