Everything we build at The AI Collective is for YOU, our community. We're gearing up to launch more content, more virtual events, and more opportunities for our members to connect.

I want to hear from YOU directly. Please take 2 minutes to complete this quick survey for a chance to win $100 in Codex Credits. The more people who answer, the better we can build the community that YOU need.

-AJ Green, Executive Director

One safety story a week, pressure-tested for what's actually happening underneath the surface.

By Lindsay Gross

You would never type your symptoms, your salary, and your marriage into one Google search. You've probably told a chatbot all three this month, in full sentences, because the format invites it and the chatbot answers like it cares. A search is a few self-conscious words. A chatbot session is the stuff you wouldn't say out loud. In a KFF tracking poll, about a third of adults said they'd turned to AI for health advice and one in six for mental health, many because it felt more private than asking a person.

Then a judge redefined what "private" means. In January 2026, District Judge Sidney Stein affirmed an order requiring OpenAI to hand over 20 million de-identified ChatGPT conversations to the plaintiffs in the New York Times' copyright suit. In a separate order in the same case, the court directed OpenAI to preserve user chats even after people deleted them. The delete button, it turns out, is a request a company can be ordered to ignore. If you build on top of these models, that is not a story about OpenAI. It is a preview of what your own logs are.

A therapist, a lawyer, or a doctor hearing what users tell a product would be protected by privilege, a legal wall that holds even in court. Your logs carry no such thing, which is exactly why a judge could order twenty million of them produced. Sam Altman said so himself: there is no legal confidentiality for what people tell ChatGPT. The intimate things users pour into your app are, in the eyes of the system, ordinary business records, subject to subpoena, retained on a schedule they will never see.

So here is the test. Before you open a new chat today, ask what you'd be fine seeing read back with your name near it. Treat the box like what it is: a record, not a confessional. The delete button is a request, not a promise. The thing you typed to feel less alone is sitting on a server with a retention schedule you'll never see, and a court has already shown it can reach in. Decide what's worth that before you hit enter.

Our SF chapter was proud to collaborate with ThinkingAI on their Agentic Growth Dinner this week. Following the Databricks Summit Expo, roughly 20 leaders gathered for an honest conversation about where AI actually stands inside real organizations. The group got specific about where AI and agents are delivering impact, and they were just as candid about where they're still falling short. 

The Sacramento chapter tried something new this week with a night on building a brand using AI, held at Frequency Coworking in Rancho Cordova. The premise was blunt: most people reach for AI to move faster, and speed without strategy just gets you lost quicker. So the room went past logos and quick content into the parts that decide whether a brand actually lands, positioning, messaging, and the trust you earn with an audience over time.

Jesse Delgado, who organizes the chapter, said the crowd kept circling one worry, how to grow a business with AI without sounding generic, or as he put it, cringe. With founders, marketers, and coaches all in the same room, the answers came from very different starting points. Judging by the night, the appetite for getting this right is real.

The AI Collective just ran the first edition of Lunch & Hack, a new biweekly online series with one rule that sets it apart from a webinar: you don't only sit through a talk, you build something while you are there. Each session pairs a short walkthrough of an emerging topic with a live group hacking block, then a show-and-tell at the end.

The opener took on how AI agents handle memory, using Claude Code and the RoBrain.dev project as the worked examples. Adelina Martiniuc, who co-hosts the series, described the format as a way to bring builders together to learn and get hands-on at the same time. The recording is already up, and the next session digs into one of the harder problems in agent design right now, giving agents a context graph they can actually remember and reason over.

Deploying multilingual voice AI requires solving structural engineering problems that standard text interfaces mask. Traditional automated speech recognition (ASR) pipelines convert speech to text for intent classification, but this two-step process discards critical paralinguistic data like tone, cadence, and urgency. Furthermore, acoustic models trained on standardized speech consistently fail against regional accents, colloquial dialects, and culturally divergent expressions of sentiment.

Overcoming these bottlenecks requires moving beyond off-the-shelf scores to build culturally calibrated evaluation frameworks, localized synthetic testing datasets to bypass voice PII restrictions, and latency-optimized hosting architecture.

Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.

For all inquiries, send us a note at [email protected].

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.