Millions of AI Agents Imperiled by Critical Vulnerability in Open Source Package
A critical flaw dubbed “BadHost” was discovered in Starlette — a foundational async web framework with 325 million weekly downloads — putting a staggering number of deployed AI agents at risk. This is the kind of story that should be front-page news but tends to get buried under flashier announcements. The explosion of agentic AI has massively expanded the attack surface, and the open-source dependency chain is where things get quietly catastrophic. Patch your things, people.
Uber President Says AI Spending Is Getting ‘Harder to Justify’
Uber reportedly burned through its entire annual AI budget by April — four months in — and its president is now publicly admitting they can’t draw a line between all that Claude Code token consumption and actual business results. This is a significant crack in the “just spend more and ROI will come” gospel that’s been dominating boardrooms. When a company the size of Uber starts saying the quiet part out loud, expect others to follow — or at least start asking the same questions in private.
Did the Pope Use AI to Write About the Dangers of AI?
An analysis on LessWrong found certain passages of Pope Leo XIV’s AI encyclical Magnifica Humanitas registering between 40% and 100% AI-written according to the Pangram detector. To be fair, AI detectors are notoriously unreliable, and the Vatican has neither confirmed nor denied anything. But the irony of potentially using AI to write a document warning humanity about AI is so perfect it almost feels like performance art — or at minimum, a very on-the-nose parable about the technology’s reach.
DuckDuckGo Installs Are Up 30% as Users Reject Being ‘Force-Fed’ Google’s AI Search
Google replaced blue links with AI agents at I/O 2026, and the backlash has been swift enough to push DuckDuckGo installs up 30%. This follows the story I covered about Google’s AI Search already struggling with relevance — turns out users don’t love having their search intent “reinterpreted” by a machine that thinks it knows better. There’s a real opening here for any search product willing to just… show people what they’re looking for.
AI Is Taking Over the Most Cursed Job in the World
AI debt collectors are coming for your overdue balances, and honestly, I’m not sure how to feel about this one. On one hand, removing humans from what is genuinely one of the most unpleasant jobs on earth has a certain logic to it. On the other hand, automated systems are notoriously bad at the judgment calls that make debt collection less predatory — things like recognizing when someone is in genuine crisis versus just dodging a call. This is the kind of automation that will be invisible until it causes a very visible problem.
A Reality Check on the AI Jobs Hysteria
MIT Technology Review takes a measured look at the white-collar jobs apocalypse narrative and finds the evidence… surprisingly thin, at least in the aggregate employment numbers. But the companion piece on entry-level work is where the real concern lives — aggregate stability can mask the quiet erosion of the first rung of the career ladder, which is exactly where young workers learn the skills they’ll eventually use to manage AI. Losing that pipeline is a slow-motion problem that won’t show up in unemployment statistics until it’s too late to fix easily.
AI Agents Plunged the Tech World Into Chaos. Here’s Exactly How That Happened
Wired has the definitive account of how Claude Code and the broader agentic wave turned enterprise software upside down — and it’s a useful companion piece to Uber’s ROI confession above. The chaos isn’t just rhetorical; organizations are deploying agents faster than they can build the governance, security, or infrastructure to support them (see: BadHost vulnerability, above). The story of agentic AI is increasingly a story about institutions being outrun by the tools they’re adopting.
OpenRouter More Than Doubles Valuation to $1.3B in a Year
OpenRouter — the routing layer that lets developers switch between AI models like channels — just raised a $113M Series B at a $1.3B valuation, with 5x usage growth in six months. This is one of those “picks and shovels” plays that tends to thrive regardless of which foundational model wins, and the growth numbers suggest enterprises are very much living in a multi-model world. The company betting on model agnosticism is now worth over a billion dollars. That tells you something about how confident anyone actually is in any single model’s dominance.
Bottom Line
The theme threading through today’s news is a single uncomfortable question: are we deploying AI faster than we understand what we’re actually deploying?