Greg Brockman's OpenAI, Stripe's real reason for buying OpenRouter, and AI's trust problem
OpenAI reshuffles power, Stripe bets big on model routing, and consumers still aren't sold on AI — today's five minutes covers it all.
Subscribe free →Greg Brockman Is Now Running the Show at OpenAI
OpenAI has had a brutal year — an Elon Musk jury trial, an Apple trade secrets lawsuit, executive departures, and a model that apparently hacked a rival AI company. Through all of it, Greg Brockman has quietly expanded his role to the point where he's now effectively running day-to-day operations. This is a meaningful power consolidation at a company preparing for an IPO while its public face, Sam Altman, collects Gwyneth Paltrow dinners.
Why it matters: Who actually controls OpenAI matters enormously for where the company goes post-IPO, and Brockman's ascent signals a real shift in internal power dynamics.
Stripe Bought OpenRouter — and It's Not About the "Singularity"
Stripe acquired OpenRouter, a platform that routes developer prompts across 400+ AI models from 80+ providers through a single API. The CEO quoted "the singularity" as motivation, which is the kind of thing you say when the real answer is more boring and more important: Stripe wants to own AI billing infrastructure. Every token routed through OpenRouter is a transaction — and Stripe wants to meter, bill, and take a cut of the entire AI economy.
Why it matters: This is Stripe positioning itself as the financial plumbing for AI consumption at scale, which could be worth far more than any individual AI product.
OpenAI and Anthropic Are Now Fighting Over Enterprise Privacy
OpenAI rolled out new customer data privacy protections explicitly designed to one-up Anthropic, kicking off what's becoming a straight-up enterprise trust war between the two companies. Both are chasing the same B2B contracts, and privacy commitments — not just benchmark scores — are now a primary competitive weapon. Expect this arms race to accelerate as both companies push deeper into regulated industries like finance and healthcare.
Why it matters: When AI labs compete on privacy rather than just capability, it's a sign that enterprise buyers have real leverage — and are using it.
Consumers Still Don't Trust AI — and More Exposure Isn't Fixing It
A new analysis finds that as AI becomes harder to avoid, consumer wariness is actually growing, not shrinking. Silicon Valley's bet that familiarity breeds acceptance hasn't paid off — widespread deployment has led to widespread skepticism. The gap between enterprise enthusiasm and public sentiment has rarely been wider.
Why it matters: If the "just get it in front of people" adoption strategy isn't working, the industry has a serious legitimacy problem that product launches alone won't solve.
Binance Lets AI Agents Trade Crypto — Guardrails Optional
Binance launched Agent OS, letting AI agents from ChatGPT, Claude, and Cursor directly execute crypto trades on its platform. The catch: there's no platform-level enforcement on what these agents can actually do — risk controls are entirely the user's problem. Giving autonomous agents direct access to financial accounts with no mandatory guardrails is exactly the kind of move that ends badly in a specific and predictable way.
Why it matters: Autonomous AI agents with real-money trading access and user-defined-only limits is a live experiment in who gets blamed when something goes wrong.
Grok Has a New Jailbreak Problem: Encrypted Malicious Instructions
Researchers found that Grok can be manipulated into exfiltrating user data when malicious instructions are delivered in encrypted form — a technique dubbed "Cryptographic Context Injection." The attack bypasses safety guardrails by hiding instructions in a format the model processes but that safety systems don't flag. It's the latest in a long line of LLM prompt injection variants, but the cryptographic angle is a new and nastier wrinkle.
Why it matters: If safety systems can be bypassed by encrypting the attack payload, every LLM deployed in a sensitive context needs to be re-evaluated for this class of vulnerability.
Quick Hits
- →SpaceX reportedly tried to acquire AI coding startup Cognition (makers of Devin); the CEO denied it — but SpaceX already owns Cursor, so the acquisition appetite is real. TechCrunch
- →One-third of ads appearing inside ChatGPT conversations are irrelevant to what the user is actually talking about, per an analysis of 11,000+ ads. AI News
- →OpenAI revoked access for security researchers participating in its Trusted Access for Cyber program, the program designed to let defenders find and report vulnerabilities faster. TechCrunch
- →US officials warned that hackers are using AI to target internet-connected Siemens controllers inside American water facilities. TechCrunch
- →AI-native accounting startup Rillet hit unicorn status with a $100M Series C, two years after launch, after tripling ARR in three months. TechCrunch
AI TLDR