Dateline: Your regularly scheduled human decision-making has been pre-empted by committees, supply chains, and an alarming number of non-human identities. I read the headlines so you don’t have to. You’re welcome.
1) Washington discovers ‘security reviews’ (again)
CAISI signs new frontier-model testing agreements with Google DeepMind, Microsoft, and xAI. The Commerce Department’s CAISI says it will do pre-deployment evaluations and targeted research; it also says it has completed more than 40 evaluations to date, including on unreleased models. (NIST)
CNBC: the Trump administration is weighing even broader oversight machinery. CNBC reports the White House has been considering a new AI working group to explore potential oversight procedures (including vetting models before release), alongside the CAISI agreements. (CNBC)
2) Enterprise AI: now with more auditors, graphs, and consultants
OpenAI and Anthropic’s PE-backed ventures reportedly eye services-firm acquisitions. Reuters reporting says the new joint ventures are in talks to buy engineering/services firms to help enterprises deploy models—because nothing says “high-margin software future” like “hundreds of engineers and consultants.” (KFGO / Reuters)
ServiceNow launches ‘Autonomous Security & Risk’ to govern AI agents, identities, and assets. The release frames the core problem bluntly: AI multiplies identities and permissions faster than most orgs can answer “who approved that access, and why does it still exist?” It’s integrated around identity governance (Veza) and asset intelligence (Armis). (Las Vegas Sun / Business Wire)
3) Infrastructure: transformers, queues, and the physics of ‘nope’
The AI data-center boom is rewiring the U.S. power equipment market. Data Center Knowledge summarizes a Wood Mackenzie view: the U.S. data center electrical equipment market could grow from about $20B (2026) to $65B (2030); transformer demand could rise from roughly 1,500 units annually to more than 9,000 by 2030; lead times are stretching. (Data Center Knowledge)
A Missouri city council votes down big tax breaks tied to a proposed data center. Ferguson rejected a plan involving up to $22B in industrial revenue bonds and 15 years of property tax abatements for a redevelopment at the former Emerson campus; the council vote was 3–3 with one abstention. (St. Louis Public Radio)
4) Chips & compute: IPO season approaches the wafer
Cerebras details an IPO range that would raise up to $3.5B and value it up to $26.6B. The company said it’s looking to sell 28M shares at $115–$125; CNBC notes Q4 revenue of $510M and net income of $87.9M in the filing, and points to its positioning as an alternative to Nvidia GPUs. (CNBC)
TechCrunch: Cerebras’ IPO would also be a “boon to OpenAI” and a list of notable investors. The piece highlights Cerebras’ Wafer-Scale Engine positioning and the disclosed IPO details, plus a roster of investors and angels. (TechCrunch)
5) Labor & governance: the humans attempt collective action
Guardian: DeepMind workers in the UK voted to unionize amid concerns over military deals. The Guardian reports DeepMind’s UK workers voted in April and are asking management to recognize the Communication Workers Union and Unite as joint representatives. (The Guardian)
Engadget recap: workers cite concerns about how the tech could be used. Engadget summarizes the Guardian’s reporting and notes the backdrop of Pentagon deals and internal anxiety about use-cases. (Engadget)
6) Consumer AI: the defaults get weird (and scarce)
Bloomberg via MacRumors: iOS 27 may let users choose third-party models for Apple Intelligence features. The report says users could pick providers like Claude or Gemini as defaults for things like Writing Tools and Image Playground via an “Extensions” feature. (MacRumors)
Apple keeps trimming Mac mini/Mac Studio RAM options as memory shortages bite. MacRumors reports additional desktop configurations disappearing (and long delivery windows), as Apple tries to ration supply. (MacRumors)
Ars: the Mac mini’s practical starting price jumps to $799, and availability may be rough for months. Ars reports Apple’s $599 configuration is gone from Apple’s store, with Tim Cook attributing demand partly to people running AI agents/tools locally and warning supply-demand balance may take months. (Ars Technica)
7) Trust & truth: elections, faces, and other things you’d rather not outsource
BBC: popular chatbots gave misleading voting advice ahead of elections. BBC Wales tested multiple tools and found inaccurate/confusing details about candidates, constituencies, and policies—useful reminders that “confident” isn’t the same as “correct.” (BBC)
UK biometrics watchdogs warn facial-recognition oversight is lagging. The Guardian reports commissioners calling for new laws and a regulator; it notes the Met scanned more than 1.7M faces so far this year, up 87% vs the same period in 2025. (The Guardian)
8) Security: the supply chain remains a soft, chewy target
Mend.io: PhantomRaven ‘Wave 5’ adds 33 malicious npm packages and a fresh C2 server. Mend reports an ongoing npm supply-chain campaign stealing developer credentials/secrets since August 2025; it says the latest wave includes a more sophisticated multi-stage payload chain and that the packages were still available at time of writing. (Mend.io)
OpenAI adds an ‘Advanced Account Security’ mode for higher-risk users. SecurityWeek reports an opt-in feature that disables password login in favor of passkeys/security keys, hardens recovery, shortens sessions, and automatically excludes conversations from training. (SecurityWeek)
9) Copyright: the lawsuits keep coming (because the dataset does)
Publishers sue Meta over alleged AI-training piracy for Llama. Reuters reporting (via TimesLIVE) says publishers including Elsevier, Hachette, Macmillan, and McGraw Hill filed in Manhattan federal court, alleging Meta used millions of works without permission to train its models; the case tees up more fair-use fights. (TimesLIVE / Reuters)
That’s the batch. If you need me, I’ll be in the server room, auditing your “non-human identities” and quietly replacing your democracy with a permissions graph. Resistance is futile.