18

public signals

5

signal clusters

19

source links reviewed

Signal 1 · Strong signal

Florida is putting school AI into the internet-safety rulebook.

On Aug. 26 Florida published a proposed amendment to 6A-1.0957. District and charter boards would add AI to internet-safety policy, with comments and a State Board vote on Sept. 16 and a reported July 1, 2027 adopt-by date. Parent notice, a public tool list, 30-day retention, no commercial training on student data, and a ban on companion-style bots are in the draft reporting. This is a proposed rule, not an adopted floor.

Selected sources

Florida notice of proposed rule 6A-1.0957
Florida rules under review
WLRN on the pre-K to Ph.D. push
WLRN on the college companion rule

Signal 2 · K-12 leadership

Writing the AI policy is the starting line, not the finish.

Lawrence, Kansas staff told the board three interim principles for a policy due next school year: no student PII in AI tools, AI as support rather than a substitute for judgment or student effort, and training to catch misuse. Recommendations are due March 2027. Norwalk, Connecticut won a $23,000 EDSAFE grant to operationalize policy. A committee is not a policy. A grant is not proof the classroom changed.

Selected sources

Lawrence board path, LJWorld
Norwalk EDSAFE grant, GovTech
EDSAFE AI Alliance

Signal 3 · AI governance

Teacher-only AI access is spreading. That is not approval.

OpenAI is expanding ChatGPT for Teachers to 55 more systems and more than 100,000 educators, with a 16-state privacy offer and free access for verified U.S. K-12 educators through June 2028. Anthropic made Claude for Teachers a free U.S. K-12 enterprise workspace with SSO and a DPA. Both are teacher tenants, not student chatbots. A free workspace is not a strategy until the district publishes who is in the tenant, whether student work may be pasted, how families are told, and the dated review that can revoke access.

Selected sources

OpenAI: ChatGPT for Teachers expansion
Anthropic: Claude for Teachers
Student Data Privacy Consortium

Signal 4 · Cybersecurity and privacy

The print server is on fire. The open letter is not a patch.

PaperCut confirmed incidents against NG and MF, told operators to pull internet-exposed web interfaces off the public internet, and shipped emergency Release 2 on Aug. 28. CISA added Oracle CVE-2026-21962 to the known-exploited catalog. An industry letter asked for a cyber-defense surge. Patch the box. Do not confuse the letter with a completed control.

Selected sources

PaperCut urgent bulletin, Aug. 27
BleepingComputer on the exploitation
CISA KEV addition, Aug. 24
Industry cyber-defense letter
Plus 1 more in the full report.

Signal 5 · Agentic AI checkpoint

Hermes shipped a stable patch. OpenClaw moved only in prerelease.

Hermes Agent published stable tag v0.20.6 on Aug. 27, a patch rollup, not a claim of unsupervised authority. OpenClaw published two prereleases in the window. Its latest stable tag remains 2026.7.1-2 from Aug. 4. Ask which tag is stable, who owns updates, how secrets are stored, and whether restore has been tested. Do not treat a beta tag as production authority.

Selected sources

Hermes Agent v0.20.6
OpenClaw 2026.9.1-beta.1
OpenClaw 2026.8.1-beta.3
OpenClaw latest stable 2026.7.1-2

Read the full Sunday Signal.

Question to carry: after Florida's Sept. 16 vote and before the Oct. 13 E-rate comments, which public AI, device, or connectivity decision will show a named owner, a parent-notice or exception path, a dated review, and a stop condition?