The brief · 9 July 2026

The AI brief, 2026-07-09

Finance and healthcare keep automating their most critical pieces, while OpenAI finally opens the floodgates after a week of throttling. Five stories, one common thread: AI is moving beyond the chatbot to make decisions on your behalf.

2 min read J / K to navigate

JPMorgan Is Testing AI Agents That Rebalance Your Portfolio Solo

A twenty-year backtest is still just a backtest. But if you're already trusting a robo-advisor with your investments, this kind of research is a preview of what lands in your banking app within two or three years. Whether it holds up in live markets as well as it does in hindsight is still an open question.

practitioners › The system beats a classic 60/40 portfolio by 0.7% annually, with lower volatility, across backtests spanning two decades.

Bloomberg →

Two AI teams are now running medical research start to finish

These systems have moved well past summarizing papers. They form hypotheses, propose experiments, analyze results, and chain into the next round, with a human still holding final say on decisions. That changes how fast a lab can chase down a drug lead.

practitioners › Google DeepMind's Co-Scientist and FutureHouse's Robin targeted acute myeloid leukemia and dry AMD, detailed in two papers published in Nature on July 9, 2026.

Nature →

OpenAI opens up GPT-5.6 to everyone, with a choice of three models

After limited access reserved for Washington in late June, GPT-5.6 is now rolling out to everyone in ChatGPT, Codex and the API. You no longer get just one model, you get three, picked according to what you're doing: deep reasoning, everyday work, or fast output without inflating the bill.

practitioners › Sol (flagship), Terra (balanced), Luna (budget), available since July 9, 2026 on ChatGPT, Codex and the API.

OpenAI →

Meta launches Muse Spark 1.1, built to actually do your tasks for you

This is an agent-focused multimodal model: it reasons, calls tools, and writes code, rather than just chatting. If you're looking to automate somewhat complex workflows instead of just asking questions, this is the type of model worth evaluating.

practitioners › Meta is also opening a preview of its Meta Model API on July 9, 2026, with claimed gains in tool use, coding, and multimodal understanding.

Meta AI →

Brussels Grades Its Own Homework on Generative AI Labeling

The Commission has just ruled that its own voluntary Code of Practice on AI-generated content transparency is good enough to satisfy Article 50 of the AI Act. In practice, if you're using ChatGPT or Midjourney to produce public-facing content, the safeguards still rest on companies opting in voluntarily rather than any strictly enforced, independently verified requirement.

practitioners › The opinion was issued on July 9, 2026 following AI Board review; Article 50 of the AI Act remains the legal benchmark for labeling AI-generated content.

European Commission →