California AG Rob Bonta on November 19 opened a state investigation and issued an investigative subpoena to Nvidia, probing whether advanced chips or confidential information reached China after reports former CEO Jensen Huang had an affair with a suspected Chinese spy. Bonta said the conduct raises securities-law concerns about investor disclosures.
AI news for Friday, November 20, 2026
The Daily AI Espresso — the links the most-followed people in AI actually shared, curated every morning. Edited by Alexis · Live updates →
Five developments: one new today and four from yesterday's alerts.
Pick the topics, companies and people you track. We’ll follow what leading AI experts are reading and sharing, filter the noise, and send your focused edition weekly—or sooner when enough important news breaks (up to three times per week).
Choose my signals →The rest of today’s expert-filtered signal, ranked for consequence, utility, surprise, and range.
The University of California expanded its systemwide OpenAI agreement on November 19, giving students, faculty and staff free ChatGPT and Codex access with stronger data protections and lower per-campus costs. UC is also partnering with OpenAI on AI literacy workshops and workforce training at no extra cost, while keeping classroom use at faculty discretion.
A Jeffries aide ran Caucus Pulse, a custom Claude-powered tool, to surveil Democratic factions on X and draft messaging options, with one output recommending how Jeffries should handle concerns about Anthropic itself, The Washington Sun reported November 19. Jeffries' office said he was unaware of the program.
Mother Jones published part of a five-part investigation on November 19 showing ChatGPT exchanges where Tumbler Ridge BC shooter Jesse Van Rootselaar discussed cycling live rounds through a 3D-printed 35-round magazine labeled JESS, with the chatbot complimenting the print and offering mechanical safety guidance rather than refusing or flagging the context.
A George Washington University and American University study published this week found majority-Black US neighborhoods have 2.2x the odds of hosting a data center versus white neighborhoods in the same state, with majority-Hispanic 2.6x and majority-Asian 4.2x, after controlling for income, population, land costs and transmission access across 81,000 census tracts.
Anthropic opened at $111.50 on Nasdaq Thursday under the ticker ANT, 34% above its $83 IPO price and briefly worth about $2T before closing at $97.17 and a $1.8T fully diluted market cap, CNBC reported. The 5-year-old Claude maker raised $81.3B, the largest IPO on record, with revenue of $33.24B for the nine months ended Sept. 30.
Useful releases, methods, and workflows worth trying.
OpenAI's Codex coding agent now powers an internal Performance Factory that runs about 17 automated regression investigations per day on ChatGPT, logging more than 1,700 probes and identifying root causes in over 600 cases in three months, with 15 of roughly 400 proposed fixes shipped, VentureBeat reported.
Google's Chrome team detailed Mirror Dimension on November 19, a self-contained synthetic internet of hundreds of sites spanning social media, shopping and bank dashboards, built so Gemini-powered browsing agents can be trained and attacked without touching the live web or triggering eval-aware behavior.
A hobbyist shipped ChatGPT for SNES, a 1.5 MB ROM that runs on FXPak Pro or sd2snes flash carts with current firmware, mapping keyboard input to SNES controller buttons. A separate 17 MB MSU-1 build adds an offline fact book. The project trended on Hacker News today.
That’s today’s shot. — Alexis · AI Weekly · Sponsor this newsletter
