The Signal — July 24, 2026
Two of the biggest labs spent this week pushing AI deeper into daily life, one into your medical questions and one into your calendar, while a new tally shows statehouses racing to put guardrails around all of it.
OpenAI opened ChatGPT's health features to everyone, but the good model costs money
OpenAI made "Health in ChatGPT" available to all U.S. users 18 and older, months after unveiling it in January. You can connect Apple Health, medical records, and wellness apps to review lab results, prep for a doctor's visit, or make sense of your sleep and activity data. OpenAI says the connected health data won't be used to train models or serve ads.
There's a catch worth understanding. Free users run on GPT-5.5 Instant, which scores lower on OpenAI's health benchmarks than GPT-5.6 Sol, the flagship reserved for paying subscribers. OpenAI's defense is that both models beat human doctors on its HealthBench Professional test, so even the free tier clears a high bar. That framing deserves some skepticism, since benchmark scores come from artificial test conditions, and a recent radiology benchmark found that none of 16 AI models matched human radiologists, in part because the chatbots stated wrong findings with high confidence instead of flagging uncertainty.
The scale here is the story. OpenAI says more than 300 million people ask ChatGPT health questions every week, up from 230 million in January. When that many people lean on a tool for medical guidance, a paywall between them and the more accurate model is a real design choice with real stakes. The feature remains unavailable in the EU, Switzerland, and the UK, where data-privacy rules and the possibility of a high-risk classification under the EU AI Act keep it out.
Sources: The Decoder · OpenAI · TechCrunch
Claude's voice mode can now pick your model and touch your apps
Anthropic rebuilt Claude's voice mode. Instead of being locked to the older Haiku-only stack, you can now choose between Opus, Sonnet, and Haiku, and by default it uses the fastest version of whatever model you were last using in text chat. The bigger change is that voice mode can reach into Gmail, Google Calendar, Slack, Canva, and Notion, so you can ask it to move a meeting, draft an email, or spin up a doc while talking.
That tool access is what separates this from OpenAI's recent voice update, which improved conversational style but still can't reach into other apps to get work done. Anthropic is betting that a voice assistant is more useful when it can act than when it just sounds better. The tradeoff is that Anthropic didn't touch the underlying voice stack, so you probably won't notice smoother interruption handling or other conversational polish. Free users are limited to Haiku with a single connected app, and the update is in beta across platforms.
Sources: TechCrunch · The Verge · Anthropic
States have already passed more AI laws this year than in all of 2025
A mid-year report from the Transparency Coalition counts 84 new AI-related laws enacted across 27 states in the first half of 2026. That already tops last year's full-year total of 73, and several large states are still in session, so the number will keep climbing. The through-line is chatbot safety for kids: new rules on disclosures, self-harm response protocols, parental controls, and limits on features designed to keep young users emotionally hooked.
A few measures stand out. Illinois became the first state to require independent third-party safety audits of frontier models, joining California and New York in adopting oversight frameworks for the largest systems. Connecticut passed a 74-page law covering chatbot safeguards, protections for minors, whistleblower rules, and content provenance. Colorado and Maryland banned surveillance-based pricing, and Vermont established protections for neural data. With federal AI rules stalled and the Trump administration pushing to limit state action, statehouses have become where most of the actual regulating is happening. This report is a few days old, but it's the clearest snapshot yet of just how fast that patchwork is filling in.
Sources: Transparency Coalition · PRWeb · Tech Policy Press
On the Editor's Desk
We looked at a few other stories today and set them aside. Sierra's acquisition of the agent startup Takeoff was interesting but thin, with undisclosed terms and light sourcing, so it didn't clear the bar for a full write-up. A batch of follow-up coverage on OpenAI's Hugging Face sandbox breach and Senator Warner's AI legislative agenda were both things we already covered earlier this week, so we skipped the reruns. That left three: a consumer-health rollout, a product update, and a policy snapshot.
