Dear Friend,
As the Internet headlines were dominated by the severe and far-reaching impact of the PHP2Shell vulnerability (which was discovered by an LLM, of course, and affects potentially 70% of all websites), this week’s edition is a little shorter than usual. That being said, the vulnerability (and the unrelated eMudhra intermediate certificate authority compromise) show that the “Defender’s Window” – the time between the good guys finding and fixing a vulnerability and the moment the bad guys exploit it – is collapsing. With the advent of open-source LLMs, which are nearly as powerful as the latest frontier models (the Chinese Kimi and GLM class models trail Claude Fable and GPT-6 by only a few months), cybersecurity is becoming a rapidly evolving battlefield. Internet infrastructure company Cloudflare saw millions of exploit attempts within hours of the vulnerability disclosure, and (unrelated to the PHP2Shell issue), US banks are now seeing unvetted AI agents accessing their customers’ bank accounts. It’s a mess – and, more than likely, only the beginning. For anyone with any IT infrastructure exposed to the Internet (which is pretty much anyone), it is high time to perform a comprehensive security audit.
And now, this...
Headlines from the Future
Moderate Your AI Use: A newly released study by Berkeley’s Brian Christian and co-authors at CMU, MIT, Oxford, and UCLA showed that after just 10 minutes of AI help, people gave up faster and got more wrong once the AI was taken away. It’s one of many studies that point in the same direction: AI makes your brain lazy, and we have to deliberately seek out ways of working (e.g., writing by hand on paper) to combat this. Still, none of this research is longitudinal yet, so the long-term damage is a hypothesis.
AI Will Negotiate For You: Anthropic is reportedly working with lawyers to train its AI to negotiate for you. It’s another step toward AI agents that act for you on their own – AI agents can already research information, order (and in some cases pay for) products and services, and even call a restaurant to book a table. Next up, they’ll negotiate a better price for your new carpet and complain to your plumber about the shoddy work. The interesting question is how AI agents will negotiate with each other – because your plumber will have one too.
Coding Jobs Are the Canary in the Coal Mine: The nerdy side of me reads the blogs of former colleagues and well-known old-time software developers. More and more of them write that; now that LLMs have taken over their world, they are looking for new jobs (outside of coding) or retiring completely (a few are quitting the Internet altogether). The frustration is palpable – here is my former Mozilla colleague Jono’s sobering reflection on where he is at (and Jono is an excellent, highly creative software developer). These voices come from one industry, but I fear that the changes we are seeing in software development are the harbinger of things to come in many other industries. As LLMs get more capable, they will take over more work and hollow out many more professions. What might be left is a whole bunch of people experiencing an existential crisis and the necessity to decouple our identity from our jobs.
What We Are Reading
China’s Tourists Flock to Experience a Pretend Typhoon as ‘Dark Tourism’ Flourishes “Dark tourism” is booming in China, where paying to experience a fake disaster is the hottest holiday trend. And honestly, in 2026, who can blame them? @Jane
AI Investors Are Worried about This Key Metric AI companies are still growing fast, but the pace is slowing. Anthropic’s revenue rose 45% last quarter versus nearly 60% the quarter before. With stocks priced for acceleration, that deceleration is what has investors worried about a bubble. @Mafe
I Paid $60 for 2 Humanoids to Clean My Home in San Francisco If you’re curious about the economics of embodied, physical AI (or you’re just looking for a house cleaner in SF), this dispatch should interest you. @Jeffrey
The Director’s Method My dear friend, Misty Megia, wrote a book! She is by far one of the most creative people I’ve ever met, and here she brings her experience from the stage into business leadership. @Kacee
How the AI Boom Might Really End: 5 Scenarios Friend of radical Martin Alderson’s prediction on how the AI boom might end. It’s well thought through, digestible, and worth a read. @Pascal
...and now for some fun!
🍪 As someone who flies a bunch – this is a delight: Delta’s Biscoff cookies were supposed to be temporary, then passengers fell in love.
🌭 Like hotdogs? You’ll like this: ‘Der Viener Schlinger’: The meat cannon that launched 20,000 hot dogs.
What We Do When We’re Not Writing
Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.
↗ Dive into the deep end: Access our complete collection of 3,200+ radical links.