Daily AI brief

AI infrastructure costs, surveillance failures and a new Copilot model

California changes data-center cost rules, reporting exposes gaps in AI border surveillance, Grok 4.7 reaches Copilot, and estimates compare Muse’s early adoption with ChatGPT’s.

2026-09-22·Updated 2026-09-22·Primary sources shown for every item

01Policy & safety
California signs seven laws to limit data-center utility cost shifting
Governor Gavin Newsom signed seven bills intended to stop AI data centers from passing utility costs to residents, The Verge reports. The package requires the California Public Utilities Commission to introduce a new rate classification for data centers.
Why it mattersData-center operators and infrastructure buyers need to account for how utilities will assign costs under the new classification. The laws establish a policy direction, but do not by themselves provide a usable rate forecast.
Read analysis
SourcesThe Verge (opens in a new tab)
02Products
GitHub begins rolling out Grok 4.7 in Copilot
GitHub says xAI’s Grok 4.7 is rolling out in Copilot. It describes the reasoning model as a successor to Grok 4.6 designed for agentic coding and complex, multistep workflows, without specifying rollout coverage or measured results.
Why it mattersTeams evaluating models inside Copilot have another option to assess against their own coding tasks. Suitability for longer workflows will depend on observed task completion and reliability, beyond the model’s stated design goals.
Read analysis
Related directory entriesGitHub Copilot — The most widely deployed coding assistant, available in most major editors.
SourcesGitHub Changelog (opens in a new tab)
03Policy & safety
Investigation finds people went undetected near AI border towers before dying
MIT Technology Review reports cases in which people crossed areas monitored by AI-enabled US border towers undetected and later died nearby. Their bodies also went unnoticed, raising questions about detection and follow-up along the US-Mexico border.
Why it mattersAgencies procuring surveillance systems need evidence of actual detection and response, alongside coverage claims. These cases make missed detections a concrete issue for operational review, although they do not establish an overall failure rate.
Read analysis
SourcesMIT Technology Review (opens in a new tab)
04Business
Appfigures estimates put Muse ahead of ChatGPT’s early mobile adoption
Meta’s AI agent Muse recorded more downloads and daily active users in the US and Canada than ChatGPT did over the equivalent period after its mobile debut, TechCrunch reports. The comparison comes from Appfigures estimates; exact counts and the measurement window are not specified.
Why it mattersProduct teams assessing consumer AI distribution have an early adoption signal that includes usage as well as downloads. It supports closer examination of Muse’s audience, while leaving retention and longer-term engagement unresolved.
Read analysis
SourcesTechCrunch (opens in a new tab)

California signs seven laws to limit data-center utility cost shifting

GitHub begins rolling out Grok 4.7 in Copilot

Investigation finds people went undetected near AI border towers before dying

Appfigures estimates put Muse ahead of ChatGPT’s early mobile adoption