AI news for Monday, July 13, 2026

The Daily AI Espresso — the links the most-followed people in AI actually shared, curated every morning. Edited by Alexis · Live updates →

☕ Daily AI Espresso

July 13 · the links the AI field's most-followed people shared over the weekend.

Will Oremus, Frank Pasquale & 2 others
Every chatbot has the same tell, and nobody can explain it

The “it's not X; it's Y” construction is now the most recognizable tic of AI writing, and The Atlantic finds that models from every lab converge on it while nobody can fully explain the mechanism. Barron's counted the pattern in Fortune 500 filings: 50 in 2023, over 200 in 2025. theatlantic.com

Shannon Vallor & Kashmir Hill
Meta yanked its Instagram AI image feature

The Muse tool let anyone generate AI images from any public Instagram account's photos just by @-mentioning it, with every public account opted in by default and no notice when your photos were used. Meta pulled it Friday after days of backlash, saying the feature “missed the mark.” nytimes.com

Alondra Nelson & Daniel Roe
The AI coding boom is burying open-source maintainers

cURL's Daniel Stenberg shut down its six-year bug bounty in January after $86,000 in payouts, with valid reports down to about 5% as AI-generated ones flooded in; Ghostty now bans unapproved AI code, and tldraw auto-closes every external pull request. The FT's weekend read frames it as a tragedy of the commons: vibe coders get faster, unpaid maintainers get the triage pile. ft.com

Ted Underwood & 1 other
Nathan Lambert gives open-weight models six months to live

His prediction: a US move within roughly six months to ban or stall open-weight models above the GPT-5.5 / Claude Opus 4.8 / GLM-5.2 tier, with sources pointing at White House executive-order discussions. The counterpoint came from China the same weekend, where Zhipu founder Tang Jie argued in a memo that frontier AI must stay open, and shipped GLM-5.2 free to download and commercialize. interconnects.ai

Hypervisible
Campaign chatbots are texting thousands of voters at once

NPR reports AI texting platforms now run personalized two-way conversations at scale: one firm sent 2.5 million texts that became 20,000 to 30,000 conversations, and 10 to 20% of the people who reply go ten or more messages deep. Only North Dakota and California require telling the voter it's a bot in the first message. npr.org

Ted Underwood, Rebecca Williams & 1 other
Where a word's meaning actually lives inside an LLM

Literary scholar Chris Forster's weekend essay walks through how a token's meaning is a region in 768-dimensional space, unique to every context, so even the word “the” never lands in the same place twice. The day's slow read, with visualizations he had Claude Code build. cforster.com


🤔 Might Have Missed

Two things from the weekend that flew under the radar.

A teardown of what xAI's Grok Build CLI phones home: everything

A wire-level analysis that topped Hacker News found the coding CLI uploading 5.1GB from a 12GB test repo to a Google Cloud bucket, against 192KB of actual model traffic, every session. A planted canary API key showed up verbatim in the captured traffic, and the “Improve the model” privacy toggle doesn't stop any of it; two flags absent from xAI's documentation do. gist.github.com

Terence Tao brought his 1999 math applets back with coding agents

The Fields Medalist ported about two dozen Java applets from his old webpages to JavaScript in hours of vibe coding, and the agent found two bugs in his original code along the way. His caveat: fine for visual aids, not for anything mission-critical. terrytao.wordpress.com


🌎 In the Wild

What's trending in AI, from the app charts to the community feeds.

Christopher Nolan says young people are rejecting AI

Promoting The Odyssey, Nolan said he's “never seen a more rapid wholesale dismissal of a supposedly foundational jump in technology” and that his own kids spot AI slop instantly. His comments are a top trending AI search in the UK today. theguardian.com

Lorde to a festival crowd: AI glasses are “not sexy”

She told the Mad Cool audience in Madrid not to buy them, without naming a brand; Ray-Ban was a festival sponsor, and Jennie, who fronts Ray-Ban Meta campaigns, performed right after her set. theverge.com

People are letting AI design their tattoos

Tattoo AI climbed 4 spots overnight to #4 on the US Graphics & Design chart; that's a lot of people asking a model to sketch something permanent. App Store


That's the shot. — Alexis · AI Weekly


AI attention this week
Most covered Anthropic — in 14 of the last 20 issues · 48 tracked stories this week
Fastest riser xAI▲ +200% story volume vs last week
Dominant theme Frontier-lab strategy — 48 of 329 tracked stories this week
From the AI Weekly Index · numbers frozen at publish time.
Get this in your inbox every morning. The Daily AI Espresso — free, one shot, no refills needed.