AI & ML Research 24 Hours

Last updated: September 21, 2026, 2:02 PM ET

AI & ML Research

MIT Technology Review published a 15-month investigation into the US-Mexico border's "virtual wall," producing the first comprehensive map of deaths along the surveillance corridor. The reporting documents how a surveillance camera stood in plain sight as one migrant died near San Diego, raising hard questions about whether the sensing infrastructure was ever designed to trigger rescue rather than detection.

Systemic Failures

Despite billions spent on border surveillance, the system repeatedly failed to catch people before they died — including the case of José Morales Bernal, who crossed on April 8, 2022. The investigation's four recommendations outline concrete fixes, from alerting protocols to interagency data sharing. The daily newsletter summarizes the full package for readers catching up.

Data Valuation

A separate piece examines how Google offered $10M for a dying airline's operational data, using the Spirit Airlines case to frame how organizations can put a dollar figure on their own datasets. The core lesson: operational exhaust — routing logs, pricing histories, maintenance records — carries real market value long before a company winds down.

Model Engineering

On the tooling side, a practitioner walkthrough flags five scikit-learn defaults that deserve scrutiny before your next model readout — a pointed reminder that AI-generated code inherits library assumptions nobody audited. Meanwhile, V7 is using GPT-5.6 to give agents institutional memory, turning scattered company files into a context graph that agents can query, which points toward retrieval architectures built around persistent organizational state rather than per-session prompts.