Welcome to your Monday Intelligence Digest!
No one wants to talk about this… but what would you do if today’s inbox delivered the cheerful news that your company is now recording every keystroke and mouse click on your work computer to train AI?
Grab a coffee (or something stronger). We’ve got departures, debuts, distillation drama, and datacenter deals galore.
OpenAI’s Redemption Arc
The start of last week looked like another stinker for OpenAI after three major departures.
Then the release blitz began:
Monday - An expanded Memory feature called Chronicle to handle more extensive tasks.
Wednesday - Workspace Agents in ChatGPT to tackle tasks with a team of AIs & GPT-Images-2, which just took back the image-generation crown.
Everyone’s so excited, none remembers that on Friday Sam Altman quietly said he’s deeply sorry OpenAI didn’t warn police about the mass-casualty event in Tumbler Ridge a few months ago (the shooter had used ChatGPT to plan the attack).
But look over there! The real villains are Chinese Opensource labs - right?
Get the Intelligence News Digest sent to your inbox every Monday Morning!
The White House Warns Chinese Devs
China’s Opensource Community responded to OpenAI Hype with a 1-2-3 counterpunch!
DeepSeek’s V4 preview was released and open-sourced a a serious bump over previous models but still lagging local competition.
Kimi-K2.6 impressed again (now open-sourced as well).
But both were topped by Alibaba’s Qwen 3.6-max-preview.
The White House has a warning, however….
This official Memorandum on Adversarial Distillation Attacks put Chinese developers on notice for allegedly stealing American AI secrets - basically using tricks to understand how leading closed-source AIs work and copying them…
Putting frontier model capabilities out for free while closed-source American companies are raising prices.
Anthropic’s Post-hype Comedown
The U.S. Government also seems to be settling its dispute with Anthropic.
Despite Anthropic’s recent popularity surging because they “stood up” to the Pentagon over the use of their technology in war.
While they garnered some enthusiasm for their Claude Design product, it didn’t last.
They quietly pulled Claude Code from some new subscription plans - but only from some users, “to test it out”.
Wild reports landed mid-week that the big scary Mythos model (the one only a few companies and the government were allowed to work with) was leaked by a group of users on Discord who simply guessed the URL.
On the brighter side, they’re easing compute constraints with a 5 GW deal with Amazon and another 5 GW deal with Google. (AI datacenters are now making up 7% of total U.S. power demand.)
Google, meanwhile, is making its own deals to be used in classified Pentagon operations.
They unveiled two new AI-optimized chips, partnered with Marvell to build two more, and created a special team to get in on the AI coding game currently dominated by OpenAI and Anthropic.
Our favourite AI Tools this week!
Wispr Flow: Effortless voice-to-text anywhere for developers & creators
Kite: Turn screen recordings into Apple-style demos
Anything: Your easy peasy personal AI App Builder
Musk’s Empire Expands
But don’t count out … SpaceX-X-xAI from the coding race!
Big moves this week to acquire small but MIGHTY AI coding crew, Cursor AI. A $10B deal to share resources with an option to purchase for $60B if all goes well. (Sucks for early Cursor investors caught up in the crypto-fraud saga.)
Grok 4.3 has gotten some love, with a reminder that their 10 T-parameter model - already in training - will roll out as Grok 4.8.
Musk also shared info on the Terafab project during the Tesla Q1 earnings call. While SpaceX (xAI) will focus on scaling chip and satellite production for a mildly ambitious 1 TW/year of compute in space, Tesla has already broken ground on the Research Fab at its Giga Texas campus - The same place where the steering-wheel-free Cybercab is now rolling off the assembly lines.
Meta’s Surveillance Shake-Up
Meta casually told employees this week that they’re going to track every click and keystroke to train AI.
Top-rated comment on the internal memo: “This makes me super uncomfortable. How do we opt out?”
Answer: There is no option to opt out”
Unless, of course, Meta decides to opt them out. They’re also laying off 10% of their staff next month.
Get the Intelligence News Digest in your inbox every Monday Morning!
That’s all for this week!
To catch the news as it happens be sure to follow us on your favourite platform!