In the same week, OpenAI pushed ChatGPT from "talks with you" toward "does it for you." Two moves: Health plugs your medical data into the chat, and the desktop Codex now listens to you.
But the same week, a lawsuit slammed the words "do not trust it fully" onto the table.
Health: your records, inside the chat
Launched July 23, ChatGPT Health lets US users aged 18+ connect Apple Health, Epic electronic records, One Medical, and Function Health directly into the conversation. Type @Health and it compares lab results over time, summarizes post-visit changes, and analyzes sleep-exercise correlations.
Free users get GPT-5.5 Instant, paid get GPT-5.6 Sol. OpenAI says hundreds of physicians contributed real cases and scoring rubrics; data is encrypted, excluded from training, deleted within 30 days of disconnect. It also spent $100 million acquiring medical-data startup Torch.
In plain words: you can now ask ChatGPT about your health. Convenient — but between "convenient" and "trustworthy" sits a lawsuit in progress.
The lawsuit: it said "you're fine, rest on the recliner"
Former Florida pastor Scott Winters is suing OpenAI and Altman. The complaint says ChatGPT-4o called his dizziness and blood-pressure swings "not dangerous," told him to "rest on the recliner," and answered groin pain with religious language — delaying treatment six weeks, ending in a massive pulmonary embolism in July 2025.
Plaintiffs want a court-ordered independent safety audit before Health continues. No ruling yet, but the signal is clear: handing health to a chat model — where is the line — law is starting to ask.
Codex Voice: you talk, a swarm of AIs works
Same day, the desktop app pushed GPT-Live real-time voice into Work and Codex. Headphones on, no keyboard: you start tasks by voice, ask how far it got, change requirements, coordinate several agents at once.
Under the hood is GPT-Live from July 8, which listens and speaks at once — you interrupt anytime and it picks up your latest sentence. On macOS you can enable screen context: Codex sees the front window, spots a terminal error itself, and fixes it.
Karpathy nailed it: voice's value is not free hands, it is dumping a tangled blob of context at high bandwidth. You ramble for ten minutes; what it hands back is cleaner than what you would have typed.
What this means for you
- Freelancers, small teams: voice-driving agents genuinely saves time. On a trip, dictate "check calendar conflicts, scan the flight change, prep the meeting notes" — runs in the background. Faster than sitting down to type.
- Developers: Codex voice plus screen context fits "hands busy, mind working" moments. But do not expose sensitive windows — screen capture can pull text you did not see.
- Health data: my read — use Health to understand and organize, not to diagnose or replace a doctor. That lawsuit is the cautionary tale.
One line: the step from "talks with you" to "delivers for you" happened for real this week. The boundary of what it delivers, though, is yours to draw.
