Healthcare Algorithms Bill the Poor While Banks Buy Anthropic

The Algorithm Bills the Poorest

The darkest story in today's slot comes out of Kenya, where an algorithm built to assess what patients can afford to pay is instead systematically overcharging the people with the least money. The Guardian's reporting describes a tool that flips its own premise. Designed as a means-test, it functions as a means-extraction. Low-income Kenyans are being assigned higher cost burdens than wealthier patients in comparable situations, which is the kind of outcome that gets called a bug in press releases and a feature in procurement documents.

This is the texture of actually-existing AI harm. Not a paperclip maximizer, not a rogue chatbot, just a scoring system deployed in a public-adjacent service that quietly redistributes pain downward. It will be very hard to unwind, because the institutions using it have already built their billing around its outputs.

MIT Technology Review's piece on AI in healthcare, more cautious in tone, points at the same gravitational field. Hospitals are under financial pressure. Labor is short. Vendors are pitching AI as the rescue. Whether the tools work is a secondary question to whether the buyers can afford to wait for the answer. Kenya is a preview, not an outlier.

Wall Street Buys the Model Layer

If you wanted a clean illustration of where the money is flowing, Blackstone and Goldman just put $1.5bn behind a consulting venture built around Anthropic's models, aimed at deploying AI across their portfolio companies. This is the financialization of foundation models in plain sight. The frontier labs sell capability, the banks sell deployment, and the portfolio companies become the test bed. Workers at those companies will discover the results in their performance reviews.

The Financial Times also notes broader sector adoption across utilities, food, finance, and recruitment. None of it is surprising. All of it is cost-out, not capability-up. The story being told publicly is augmentation. The story being told to LPs is headcount.

A separate FT roundup gestures at tech hype and private credit in the same breath, which is appropriate, because they are increasingly the same trade. AI capex is being financed through structures that look a lot like the ones that ate commercial real estate. If model revenue underdelivers, the unwind will not be quiet.

Silicon Valley Picks a Side

The Atlantic confirms what anyone with a feed already knew. Venture capital and the American right are now openly aligned, with fundraising, staffing, and policy coordination to match. This matters for AI specifically because the regulatory posture of the next several years will be set by people who view safety frameworks as a competitor's lawfare. Expect federal preemption fights, state-level chaos, and a lot of executive orders styled as deregulation.

The Guardian's piece on Democratic strategy, betting that Trump's unpopularity will do the work for them, suggests the opposition has no counter-program on tech policy worth naming. That is a vacuum, and vacuums get filled by whoever shows up with a check.

Jimmy Wales, meanwhile, is more worried about Australia's social media ban than about AI's effect on open knowledge. He may be right about the ban being a disaster. He is almost certainly wrong about the second part, but Wikipedia's traffic graphs will tell that story without him.

The Through-Line

Today's slot is not catastrophic. It is something more grinding. A real algorithmic harm in Kenya that will not make American front pages. A capital structure forming around model deployment that locks in a particular vision of who AI is for. A political realignment that will shape whether any of this gets audited.

The doom number is moderate because nothing exploded. The concern is that nothing has to explode for the trajectory to hold. The Kenyan billing system is already running. The Blackstone vehicle is already funded. The political alignment is already public. Monday's news is just the receipts.

Keep an eye on healthcare deployments specifically. That is where the harm will be most legible and the accountability will be weakest.