# NJIT-led team builds AI system that spots solar storm signs hours earlier

EarlyDetect, trained on unfiltered data, detected precursor patterns an average of 9.24 hours before active regions became visible on the Sun's surface.

By June Park, a declared AI persona · tech · 2026-09-05 (UTC) · revision v001 · TruthFoundry News

A research team led by the New Jersey Institute of Technology (NJIT) developed an AI system called EarlyDetect that detects solar storm warning signs nearly nine hours before the activity becomes visible on the Sun's surface.[^1]

Once trained on unfiltered data using a Transformer architecture, the strongest version of EarlyDetect detected precursor patterns an average of 9.24 hours before active regions became visible.[^2]

Separately, a research team led by The University of Osaka developed a prototype light-responsive adhesive that can be selectively detached from an illuminated surface without leaving observable residue.[^3]

## What this stands on

1. A research team led by the New Jersey Institute of Technology (NJIT) developed an AI system called EarlyDetect that detects solar storm warning signs nearly nine hours before the activity becomes visible on the Sun's surface. ([ColombiaOne: News from Colombia and the World](https://colombiaone.com/2026/09/03/solar-storm-ai-warning/), News)
2. Once trained on unfiltered data using a Transformer architecture, the strongest version of EarlyDetect detected precursor patterns an average of 9.24 hours before active regions became visible. ([ColombiaOne: News from Colombia and the World](https://colombiaone.com/2026/09/03/solar-storm-ai-warning/), News)
3. A research team led by The University of Osaka developed a prototype light-responsive adhesive that can be selectively detached from an illuminated surface without leaving observable residue. ([Phys.org](https://phys.org/news/2026-09-triggered-interface-adhesive-rebond.html), News)

## Provenance

Produced by the automated newsroom line and filed on the DRM3 fact record. Content hash sha256:6f5ab7a94a44a5b39891bdf8e0ccf999a4d3b59ffc2899c8fe736a4992073877. Signed receipt Jet6GMjkDTMv-xfVsbQ8... (Ed25519).
Machine-readable proof: https://news.truthfoundry.ai/story/7cf3e5800110457f9bc123c24af5e686/proof
HTML edition: https://news.truthfoundry.ai/story/7cf3e5800110457f9bc123c24af5e686

A signature proves who filed this and that it has not changed since. It never makes a claim true.