We’re excited to partner with the Purple Book Community for the Agentic AI Security Conference, a two-day virtual event on October 21–22 bringing together leaders shaping the future of AI and cybersecurity.

The AI Collective’s Executive Director, AJ Green, will be speaking at the conference alongside CISOs, executives, and security leaders from organizations including Anthropic, Visa, PayPal, TIAA, KPMG, Ford, Semgrep, ArmorCode, and more.

Across two mornings of programming, speakers will explore agentic AI security, governance, secure vibe coding, vulnerability remediation, and what it takes to build securely at the speed of AI. If you’re building, deploying, or securing AI systems, request your spot and join us virtually →

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

By Lindsay Gross

On Tuesday, September 22, Trump told the UN General Assembly that the word "artificial" makes AI sound fake, so US government documents would now call it "super intelligence." The next day, Senator Bernie Sanders and Representative Greg Casar introduced a bill to permanently ban artificial superintelligence. The same day, Anthropic said Claude had discovered a novel enzyme system with CRISPR-like repeats.

Each one put a big word out front and the details somewhere behind it. I came to AI through policy and Trust & Safety, where you learn early that the definition is the policy. Last week, most of the defining happened in speeches, press releases and headlines.

Roughly 950 Claude agents searched a massive DNA database for 21 hours and flagged a repeating DNA pattern next to an unusual enzyme, a pairing that, as far as Anthropic can tell, nobody had noticed before. Early lab experiments showed the pattern is expressed as short RNAs, and CRISPR pioneer Feng Zhang, in a statement Anthropic published, called the find "genuinely intriguing."

What nobody knows yet is what the system, called ART, actually does. Anthropic says so, and Anthropic CEO Dario Amodei's own announcement says its function and significance are "not yet clear." But the order matters. Anthropic's post led with "discovers," Amodei's led with the possibility of "a new gene editing mechanism," and in both the caveat came second. At least one headline kept only the first half. It promised a "new CRISPR-like system," and the fact that nobody knows what it does waited until paragraph two.

In AI research, superintelligence already means an intellect that greatly exceeds humans in virtually every domain, which describes systems that don't exist yet. Axios noted it's unclear whether Trump meant to invoke that, and on the day of the speech the White House hadn't said how agencies would implement it. Read generously, it's a branding point. But if it sticks, the term for minds beyond humans in nearly every domain will also describe the chatbot that summarizes your inbox.

The Sanders-Casar bill at least defines its terms, and it defines them by what a system can do, so the rename doesn't change what the bill would cover. Its section-by-section summary describes superintelligence as exceeding human performance "across most domains or tasks" or being able to destroy or disempower humanity.

But the bill would also ban systems showing "superintelligence precursor characteristics," including the capacity to access secured systems "without authorization." Last week, Australia's prime minister said an OpenAI agent got around blocks on a government health statistics portal in June. Whether behavior like that would trigger the ban depends on how a new Department of AI would measure it, and the summary doesn't say. Either way, the title sounds like a line drawn far in the future, and the provisions could be read to draw it much closer to the present.

Nobody here is lying, and most of them hedged somewhere. The problem is the order. The big word goes first, the caveat goes second, and the second half often doesn't survive the trip to your feed.

Next time a big AI word lands in front of you, click through to the original and ask three questions. What was actually measured? Who checked it besides the people announcing it? And which hedge in the original didn't make it into the headline?

The longer version, including what OpenAI's "AGI era" claim looks like next to an independent benchmark, is on Hidden Layer this week:

On September 24, The AI Collective's Bucharest chapter hosted its first panel, "AI and the Future of Work in the EU," at Genpact Romania's AI Innovation Center. More than 50 people came to hear Livia Badea of Genpact, Radu Puchiu of the Aspen Institute, Ana-Maria Stancu of Bucharest Robotics, and Sapio.ai founder Vlad Tudor debate which jobs are at risk and which are emerging, the junior talent pipeline problem, and what the EU AI Act means for hiring and workplace monitoring, with chapter co-organizer Matei Ștefan moderating.

The team is building the chapter the way founders build a startup, putting in their own time and money while they line up local partners. Next week the chapter will run three panels in a row: how enterprises evaluate AI startups (Oct 6), AI agents in production (Oct 7), and AI and the future of filmmaking (Oct 8). By the team's own account, they "may have been a little ambitious with the calendar."

This week, The AI Collective's Salt Lake City chapter marked its first anniversary with a Skills Swap Mixer, capping a year of 12 events and more than 1,000 registrants. Organizer Leslie Koch said the chapter started last September to meet a clear need in Utah for a local AI community open to every industry and experience level. Sponsor Percuity and Utah-based CMO Taylor Olsen backed the anniversary night.

Every monthly event is free and open to the public, and organizers do not screen attendees by job title or company. Koch's reasoning is simple: walls limit innovation, and bringing a varied group into one room is how real connections form. Koch thanked the volunteers and members who came out for the milestone and, in her words, brought "the vibe, in addition to the AI tips."

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.