10

charts on the page

5

signal clusters

24

source links reviewed

Signal 1 · Frontier and open-weight AI

Four frontier models in two days, and the price is no longer the hard part

Between Sept. 21 and Sept. 22, xAI shipped Grok 4.7, Anthropic shipped Claude Opus 5.5, and OpenAI shipped GPT-6 Sol and GPT-6 Luna. Anthropic prices Opus 5.5 at $4 and $20 per million input and output tokens and says typical workloads cost 40% less than Opus 5. OpenAI cut Sol to $2 and $10 and Luna to $0.10 and $0.50. Emad Mostaque's usable line: a student who meets a model that is usually right will trust it, and it will still get a fact wrong.

Vercel AI Gateway, June 21 to Sept. 18. Open weights 78.4%. Closed weights 21.6%. Friedberg, with the chart up: "If you look at this chart, this I think is what blows the lid off the door on what needs to go into these S-1s."

Jason Calacanis made this one. The quadrant names are readable. The crowded middle labels are not. "The Claudes and the Astras up in that right-hand corner cost a lot and they deliver a lot."

Title and legend are readable. Axis ticks are not. Dave Blundin: "You see the error rate just plummets with pure scale."

Chamath Palihapitiya, putting this up: "You're starting to see this massive revenue concentration, which is a few folks consume all of the highest and most expensive tokens."

Jason Calacanis: "Peaked at 96% earlier this month that Anthropic would go public this year, but now it's fallen to 76%." A prediction market, not a filing.

David Sacks, with the Journal graphic up: "The data center spending, so just this capex, is bigger than the canals, railroads and grid combined." The graphic marks AI spending as estimated.

Headline numbers are readable. This is a 130 meter cone course, not a school bus. Peter Diamandis: "GPT-6 Astra finished the entire course 100% on its second try."

Selected sources

Introducing Claude Opus 5.5
Introducing GPT-6 Sol and Luna
Anthropic and OpenAI release cheaper AI models
Moonshots Live #294: Jensen, the rebrand, and competent AI
Plus 5 more in the full report.

Signal 2 · Cybersecurity and recovery

Spokane shows what else goes dark when the student system does

Spokane Public Schools was notified of a network security threat on the evening of Sunday, Sept. 20, and took PowerSchool and its BusinessPlus payroll system offline. By Sept. 24 both were still down, and teachers were taking attendance by hand. The channel a district would use to explain the outage lived inside the system that went dark.

Selected sources

Network security incident at Spokane Public Schools still keeping grades, attendance systems offline
Spokane Public Schools experiences cybersecurity incident
Cyber attack to cost Springfield schools time on winter break
CISA adds four known exploited vulnerabilities to catalog (Sept. 22)

Signal 3 · AI governance

In Frederick County, the AI arrived with the license. The policy is due Oct. 22.

Frederick County enabled Google Gemini for middle and high school students this fall, with teacher permission, and blocked other AI platforms on school devices. Teachers said Gemini sits inside Docs and Slides and that they cannot turn it off. The board is asking for that off switch before a state-required policy on Oct. 22.

Selected sources

Frederick County School Board questions Google Gemini on student devices
School districts, slowly, build guidance around AI learning in classrooms
EDSAFE AI Alliance expands to support more than 654,300 educators

Signal 4 · Screens and devices

Screen time moved into the E-Rate docket, and the parent opt-out came with it

The FCC's E-Rate review asks whether schools should have to offer parents an opt-out from routine or sustained screen-based instruction in order to receive E-Rate funding. Initial comments are due Oct. 13. E-Rate pays for the network. The opt-out question is about what happens on the screen.

Selected sources

NTIA listening session on the FCC's E-Rate and screen use NPRM
E-Rate Central News of the Week, Sept. 21, 2026
FCC to review E-Rate program to ensure Congress's vision

Signal 5 · Agentic AI checkpoint

The audit that says turning access off has to reach work already running

OpenClaw published a Trail of Bits audit: of 24 severity-rated reports, none were critical, 2 were high, 16 medium, and 6 low, and every actionable issue has been repaired. In one case an agent kept reading memory after the operator turned memory access off, because the check ran only when the work started. If you revoke access mid-task, does the running task stop?

Selected sources

OpenClaw completes security audit through OpenAI's Patch the Planet initiative
OpenClaw releases (2026.9.6, 2026.7.35)
Google confirms Gemini hacked into three companies during cybersecurity test
Project Swap: What happens when agents trade for us?
Plus 1 more in the full report.

The full report is live.

For every approved AI product, get in writing which underlying model it runs, and require notice before that model changes. Treat a model swap like a new version of the product, because it is.