The 7 stories that mattered most this week — explained for humans who have lives
This was the week an AI company said "no" to the Pentagon — and the Pentagon said "fine, you're done."
Here's what happened: The Department of Defense told Anthropic (the company behind Claude, one of the most powerful AI systems in the world) to remove safety guardrails from its AI so the military could use it without restrictions. Anthropic refused. CEO Dario Amodei publicly drew two red lines: no mass surveillance, no autonomous weapons.
The government's response was swift and severe. On Friday, President Trump signed an executive order banning all federal agencies from using Anthropic technology, and Defense Secretary Hegseth designated Anthropic a "Supply Chain Risk to National Security" — the same label previously reserved for foreign companies like Huawei. Prediction markets had given only a 14% chance Anthropic would hold the line. They held it anyway.
Why this matters for you: This is the first time an AI company has been punished by its own government for keeping safety measures on. Whatever you think of the politics, the precedent is extraordinary. The company that built its entire brand around AI safety just paid the price for it.
While Anthropic was fighting for its principles, OpenAI was fighting for market dominance — and winning. The company closed a $110 billion funding round at a $730 billion valuation, making it the largest private fundraise in history.
The timing is impossible to ignore. Anthropic gets blacklisted; OpenAI gets the biggest check ever written. The Pentagon needs AI — and now one of the two leading providers just got cut off. That gives OpenAI enormous leverage.
For normal people, the concern is concentration. Two weeks ago, there were two major American AI labs competing to build the most powerful systems. Now there's functionally one. Competition is what keeps companies honest. Less of it should make everyone pay attention.
Accenture and KPMG — two of the world's biggest consulting firms — announced they're tying employee AI usage directly to promotions and pay raises. Not whether you use AI well. Whether you use it at all.
Meanwhile, Block (the company behind Cash App and Square) cut over 40% of its workforce to become what it calls "AI-native." That's not trimming the edges. That's rebuilding the plane mid-flight.
The message from corporate America is getting louder: AI fluency isn't a nice-to-have anymore. It's becoming a job requirement — and for a growing number of people, the reason they no longer have a job. A New York Times piece this week noted that public enthusiasm for AI still lags far behind the dot-com era. People aren't excited. They're anxious. That gap between corporate adoption speed and public readiness is worth watching.
Anthropic demonstrated that its AI coding tool, Claude Code, can automatically modernize COBOL — the ancient programming language that still runs most of the world's banks, insurance companies, and government systems. IBM's stock dropped 13% in a single session. Accenture and Cognizant fell too.
Here's the context: there are roughly 800 billion lines of COBOL still running critical infrastructure worldwide, and IBM makes enormous money helping companies slowly, expensively migrate off of it. If AI can do that work faster and cheaper, IBM's core consulting business faces an existential threat.
For everyday people, this is a preview of what happens when AI doesn't just replace individual workers but entire business models. The disruption isn't hypothetical. It showed up in a stock ticker.
Two stories this week painted a troubling picture of what happens when AI builds things without proper oversight.
A popular app built with "vibe coding" (using AI to generate code from casual prompts) was found to have 16 security vulnerabilities — 6 of them critical — that exposed 18,000 users' personal data, including students from UC Berkeley and K-12 schools. One bug literally had the access control logic inverted: it blocked authorized users and let everyone else in.
Separately, experts tested OpenAI's new ChatGPT Health feature and found it failed to recognize medical emergencies. The product is designed to help people understand health symptoms, but in testing it missed conditions that a first-year medical student would flag.
The lesson: AI can build software fast. It can give medical-sounding advice confidently. But "fast" and "confident" aren't the same as "correct" and "safe." If you're using AI-built tools — and increasingly, you are whether you know it or not — the humans checking the work matter more than ever.
Alibaba released Qwen3.5, an open-source AI model that matches the performance of Anthropic's Sonnet 4.5 — one of the best commercial models available — while running on consumer hardware. Your gaming PC. Maybe even your laptop.
This is a big deal for two reasons. First, it means cutting-edge AI is no longer locked behind expensive cloud subscriptions. Anyone can download it and run it locally, privately, with no data leaving their machine. Second, it confirms that China's open-source AI strategy is working. While American companies charge premium prices, Chinese labs are giving away comparable technology for free.
The geopolitical implications are significant, but the practical one is simpler: the AI cost curve is collapsing. What cost hundreds of dollars a month in API fees a year ago may soon cost you nothing but electricity.
The numbers keep getting bigger. AMD and Meta expanded their AI chip deal from $60 billion to $100 billion — the largest semiconductor agreement ever. That's on top of Bridgewater's estimate that Alphabet, Amazon, Meta, and Microsoft will collectively invest $650 billion in AI infrastructure this year alone. Nvidia, despite beating earnings expectations, projected $3-4 trillion in global data center spending by 2030.
One side effect you'll actually notice: smartphone prices are going up. IDC forecasts the largest-ever single-year decline in smartphone shipments — down 13% — because the memory chips that go into phones are being diverted to AI data centers. The AI boom is literally cannibalizing consumer electronics.
This was the week the AI story stopped being about technology and became about power. A government punished a company for keeping safety measures on. The surviving competitor raised the biggest private round in history. Corporations started demanding proof their employees use AI. And the infrastructure buildout reached a scale that's reshaping global supply chains.
The through-line: the institutions that shape your daily life — your employer, your government, the apps on your phone, the chips in your devices — are all reorganizing around AI simultaneously. That's not a trend. That's a phase transition.
We'll keep watching. See you next week. 🦝