and extracts the following text as a script body — the dev server then fails its dependency scan with `Expected ";" but found "non"` from the words "non-allowlisted hosts", which breaks e2e. So, what this buys, so nobody reads more or less into it: blocks external `script src=` from non-allowlisted hosts blocks injected inline scripts — each one is pinned by hash blocks fetch/XHR/WebSocket exfiltration (connect-src allowlist) blocks form-action, base-uri, object-src vectors does not stop image-based exfiltration — img-src must allow https: because event media comes from arbitrary publishers' og:image The cost of hashing is that an inline script edited after the hashes are computed is silently refused by the browser. That is why the violation reporter below exists: report-uri/report-to are ignored in a meta element and Pages serves no CSP header, so violations are caught in-page and sent to PostHog instead (src/lib/csp-report.ts, #241). frame-ancestors is deliberately NOT in this string: it is ignored in a meta element and browsers log a warning about it on every page load. It lives in public/_headers instead, ready for a CDN in front of Pages. X-Frame-Options (set there too) is the clickjacking control that actually applies today. */ }
breakthrough medium confidence

UT San Antonio Researcher Builds Self-Powered TinyML Flood Warning Station Costing $150–$220

| AI for Good

UT San Antonio's Chen Pan, assistant professor of electrical engineering, announced on September 28, 2026 a self-powered flood warning prototype that runs TinyML models directly on a microcontroller, reporting 98.82% validation accuracy without any cloud connection. Each solar-powered station combines temperature, humidity, light and precipitation sensors with four optical water-level sensors at different heights, and relays alerts over LoRa radio (about half a mile in urban areas, up to five miles in open terrain). Components cost roughly $150–$220 per station. The work, funded by a NOAA-backed Texas Coastal Management Program grant and done with Texas A&M University-Corpus Christi, targets low-cost hyperlocal flood alerts for communities lacking dense gauge networks. This is a university prototype, not yet a fielded deployment.

Media

Flooded street; UTSA self-powered TinyML flood warning research
Flooded street; UTSA self-powered TinyML flood warning research — UT San Antonio Today

Sources

View full AI for Good dashboard