Morning Edition

The Futurist

Saturday, June 20, 2026
AI & Technology Markets & Crypto Ideas Worth Keeping
S&P 500$7,500.58▲ 1.08%
Dow Jones$51,564.70▲ 0.14%
Nasdaq$26,517.93▲ 1.91%
VIX$16.88▲ 2.93%
BITCOIN$62,327.00▼ 2.94%
RIPPLE$1.12▼ 4.64%
ETHEREUM$1,690.16▼ 3.08%
DOGECOIN$0.08▼ 3.47%
01

AI & Technology

404 MEDIA

ICE Appears to Be Buying Immigrants' Tax Identifiers from a Data Broker

A $10 million procurement suggests ICE is purchasing IRS-adjacent records through a data broker to fuel deportation operations — potentially skirting a court order in the process. Senator Ron Wyden called it out directly. This is surveillance infrastructure being quietly assembled in plain sight.

SIMON WILLISON

GLM-5.2 Is Probably the Most Powerful Text-Only Open Weights LLM

Chinese lab Z.ai dropped a 753B parameter model under MIT license — open weights, genuinely competitive at the top of the leaderboard. While the US government is banning frontier models, China is open-sourcing them.

404 MEDIA

If AI Is Sentient Then So Is 'Age of Empires II'

A new paper uses formal logic to demonstrate that the criteria used to argue AI sentience would also apply to a 1999 strategy game. The point isn't to demean AI — it's to expose how sloppy our anthropomorphism has gotten.

404 MEDIA

Salesforce's Internal AI Leaderboard Has Teams Competing for Little Trophies

Salesforce built a gamified internal system ranking teams by AI adoption, complete with badges and a "click to see who 👀" shaming feature for non-adopters. Corporate AI mandates have entered their surveillance era.

A16Z

How Exa Is Building the Perfect Search Engine for the AI Era

Exa's CEO makes the case that traditional search was never designed for AI agents as users. When the query-maker is autonomous, the entire retrieval stack needs rebuilding from scratch — and that's a foundation-level infrastructure bet.

SIMON WILLISON

The Real Value of MCP Is Auth, Not Protocol

Sean Lynch argues MCP's killer feature isn't the model context scaffolding — it's isolating auth flows outside the agent's context window entirely. The idealized MCP might just be an auth gateway, full stop. Small insight, large implications.

02

X / Twitter Signal

Andrew Ng flagged something that deserves more attention: within two weeks, both the US government and Anthropic demonstrated they have unilateral power to cut off access to frontier AI. The Claude Fable 5 ban — whatever its full details — proved that a single letter can kill a model's availability overnight. Ng frames this as a sovereignty issue, and he's right to.

Ethan Mollick raised a quiet but sharp point: companies are systematically undervaluing higher-intelligence AI on tasks where cheaper models "seem good enough." If the KPI is hit either way, the delta in output quality gets ignored — until a competitor who did use the better model shows up with a meaningfully superior product.

Sebastian Raschka confirmed GLM-5.2 as the best open-weight model available today, echoing Simon Willison's take. Two credible technical voices saying the same thing on the same day isn't noise — it's signal. The open-source frontier just moved, and it moved eastward.

The Neuron flagged OpenAI revisiting 376 rare pediatric disease cases and surfacing 18 new diagnoses. Healthcare AI tends to get buried under the hype cycle, but diagnostic applications on rare diseases — where human expertise is genuinely scarce — may be where near-term impact is most concrete and least contested.

03

The Thread

Two stories this week that look unrelated are actually the same story. ICE buying immigrant tax identifiers through a data broker and the US government banning Claude Fable 5 via a letter to Anthropic are both about the same underlying reality: infrastructure — data, compute, model access — is now a primary lever of state power. Governments don't need to build tools anymore. They buy them, ban them, or pressure the companies that run them.

While the US bans frontier models, China open-sources them. That asymmetry has a compounding effect.
The GLM-5.2 release under MIT license is the sharpest illustration of this dynamic. Z.ai drops a 753B parameter model free for anyone to run, the same week the US government demonstrates it can shut down a leading American model with a letter. Andrew Ng noticed. Sebastian Raschka noticed. The strategic irony is almost too clean to be real, but here we are. The Salesforce leaderboard story is easy to mock — little trophies, emoji shame buttons — but it's actually a preview of how AI adoption gets forced inside large organizations. Not through inspiration, but through visibility and social pressure. The "click to see who 👀" feature is HR surveillance dressed up as gamification. Expect more of it. The Ethan Mollick observation connects here: companies are optimizing for KPI coverage, not output quality, which means the adoption being tracked on these leaderboards may be mostly theatrical. Busy work, benchmarked.