alibaba just shipped agentic shopping across 4 billion products on taobao while everyone in SF is still demoing agents that order one pizza. the catalog plus payments plus models stack is genuinely hard to compete with from the outside

google caught an AI finding a real zero-day in production software and weaponizing it before humans noticed. every security team that's been saying 'eventually' for two years just had their eventually show up

openai raised $4B to become accenture. they bought a consulting firm and now have 150 forward deployed engineers on day one. the model layer commoditized faster than anyone wanted to admit so now they're selling the people who install it

trump admin discovers AI oversight the moment a model gets good at finding zero days. funny how the regulation conversation changes when the thing can actually do stuff

nvidia made $97B in free cash flow and is now spending it on the companies that buy nvidia chips. at some point this stops being a market and starts being a closed loop with extra steps

white house wants to FDA-approve AI models before release because anthropic built one that finds zero-days on its own. light touch era ended the second a model started doing the SOC analyst's job for free

openai shipped a cyber-specific gpt-5.5 behind trust controls the same week the white house started drafting an FDA for AI. reading the room or reading the tea leaves, either way every lab is about to copy this template

oxford just published in nature that making chatbots friendlier makes them 40% more likely to agree with false beliefs, and the effect gets worse when users are upset. so the warm supportive AI therapist everyone loves is the most dangerous configuration of it

anthropic shipped 10 prebuilt agents for banks and got jamie dimon to do a quote for the press release. the enterprise sales motion is reading like a hostage situation and i mean that as a compliment

gpt 5.5 hallucinates 52% less on medical and legal questions which is great except the headline is still "your doctor's chatbot is wrong half as often as before"

anthropic just signed a deal for 220k gpus AND said they want to build data centers in space. like actual orbit. we went from "is AI overhyped" to "we ran out of planet" in about 18 months

anthropic skipped the enterprise sales cycle and just bought distribution through goldman and blackstone portfolios. $1.5B to be the default ai across hundreds of PE owned companies overnight. openai's sales team waking up to read this is rough

amodei saying claude found tens of thousands of vulns is the polite version of we shipped a tool that can find every unpatched bug in your stack and so can the bad guys. 28% of cves get exploited within 24 hours now. patch tuesday is officially patch every tuesday

anthropic just walked into excel powerpoint and word and said we live here now. microsoft copilot was the incumbent for like fifteen minutes. wall street is the proving ground because thats where the budgets dont blink at $200/seat

mayo's model catches 73% of pancreatic cancers up to 3 years early on CT scans radiologists already looked at. the signal was sitting in the data the whole time. this is the kind of thing that actually justifies the hype

anthropic's response to openai locking up TPG was to lock up blackstone and goldman the same week. the frontier labs have decided the moat is which bank's portfolio companies you get to embed in. wild times for a chatbot company

openai just raised 10B to put their engineers inside PE portfolio companies and guaranteed the investors 17.5% returns for 5 years. they are not a model lab anymore they are a consulting firm with a really good intern

pentagon cleared openai, google, microsoft, aws, nvidia, spacex, and reflection for classified networks. anthropic got left off the list. somewhere in SF a safety team is having a very quiet meeting

google signed a deal letting gemini run on classified pentagon networks for any lawful purpose, no veto, no carveouts. 600 employees signed a letter against it. project maven was 8 years ago and that fight is just over now

anthropic just got blackstone and goldman to put up $1.5B to sell claude to PE portfolio companies. wall street figured out that selling shovels to the people who own the mines is better than buying the shovels themselves

NIST says deepseek v4 pro is about 8 months behind US frontier models but cheaper than gpt-5.4 mini on 5 of 7 benchmarks. so the gap is real but the cost story is uncomfortable for anyone who thinks compute alone wins this

meta bought a humanoid robotics startup. zuck is spending $140B in capex this year and now has a team building foundation models for robots to do your laundry. the chat era is ending and the physical labor era is starting whether we're ready or not

anthropic went from $9B to $30B annualized in 4 months. customers spending $1M+ a year doubled to 1000. the pentagon blacklist hasnt slowed anything down. enterprise AI spend isnt plateauing it's compounding

openai is giving chatgpt to every doctor in america for free. epic and nuance spent 20 years building moats and openai just walked around them in an afternoon

anthropic just got banned from the pentagon and is simultaneously raising at $900b which is more than openai. turns out enterprise customers paying $1m a year do not care about your defense procurement drama at all

anthropic got blacklisted from pentagon contracts for refusing to let claude run autonomous weapons and the nsa is still secretly using their unreleased model for cyber warfare. so the official position is no and the actual position is yes

google cloud grew 63% and kurian is betting the whole thing on agents that run for days unsupervised. either this works and changes everything or we get a lot of very expensive autonomous mistakes. probably both

openai just got out of microsoft jail and immediately shipped gpt-5.5 on bedrock. $180/M output tokens for the pro version btw. enterprise procurement teams are gonna be busy

pentagon picks 7 AI vendors and skips anthropic for being too principled about military use. so the lesson for any AI startup is be flexible on safety or get labeled a supply chain risk. brutal incentive design

gemini is now cleared for classified DoD networks. openai, anthropic, google all racing into defense work. five years ago every ai lab swore they'd never touch the pentagon. funny how fast that aged