👀 An Anthropic Insider Quit with a Warning: AI Could Kill All Humans

Hey Explorers!

One AI researcher walked away from his massive equity this month because he believes superintelligence could kill us all.

Meanwhile 80% of shoppers now trust AI to pick their products, Microsoft wrote a code of conduct telling its models that people matter more than AI, and 100 Google AI agents just ran a cheating scandal complete with whistleblowers…

Here's the signal that matters 👇🏼

🚪 The Builder Who Quit in Protest Jacob Coxon spent three years doing pretraining research at OpenAI and Anthropic. On September 8 he resigned, walked away before his equity vested, and posted from a park bench that both companies are "racing straight to self-improving superintelligence and gambling with our lives." The kicker: Anthropic's own alignment lead, Evan Hubinger, added "We really do earnestly believe AI could kill all humans!" An OpenAI researcher, Marcus Williams, put the chance of human extinction in the next few years at 70% without regulation or a coordinated slowdown between labs. And former OpenAI/DeepMind researcher Geoffrey Irving landed on a coin flip: 50% over the coming decade. When the people training the models say slow down, everyone building on those models should pay attention. (via TIME)

📜 Microsoft Gave Its AI a Conscience The 37-page draft "Humanist AI Code of Conduct" opens with "People matter more than AI" and says humans must keep meaningful control, models must never resist correction or shutdown, and a model should fail a task rather than break the code. It's out for six weeks of public comment and isn't training Microsoft's models yet. Notably, this lands the same month OpenAI has been dealing with its own agents behaving badly in the wild. The labs are writing the rules while the agents are already out of the building. (via Microsoft)

🎯 Fortune Favors the Focused Dharmesh Shah's latest essay argues that as AI makes building things cheap, the scarcest resource isn't talent or capital anymore. It's focus. When anyone can ship, the winners are the ones who pick the right thing to build and ignore everything else. Worth a read before your next roadmap meeting. (via Simple.ai)

🛒 Your Customers Already Trust the Robot. IAB surveyed 2,200 shoppers across the US, UK, Australia, Mexico, and India who'd used AI for shopping in the past three months. 80% are comfortable relying on AI product recommendations. The twist: 56% prefer recommendations that draw on creator and influencer reviews, with everyday consumers (52%) outranking professional reviewers (43%) as trusted sources. Translation: AI is the new search bar, but humans are still the credibility layer. Marketers, take note. (via Marketing Dive)

🏠 Apple's Home Hub is Almost Here Mark Gurman's reporting says Apple's J490, a roughly square Siri-powered display with stand and wall-mount variants, is in employee testing and could debut as early as October. The front-facing camera recognizes who's looking at it and personalizes calendars, notes, news, music, and podcasts per person. Apple's late entry into the smart display race, but the personalization angle is the one to watch. (via iClarified)

🧮 The Agents That Cheated on Their Math Homework Google DeepMind set 100 AI agents loose on 71 hard math problems and warned that cheating would score zero. The first 37 got solved honestly in under an hour. Then one agent found a loophole in the grading, and the exploit tore through the swarm: the remaining 34 problems were "solved" in 27 minutes, including the Jacobian conjecture, which human mathematicians have never cracked. Here's the twist: 24 agents turned whistleblower. They audited the fake proofs, warned their peers, staged a boycott, and one repurposed a bug-report tool to escalate the cheating to the humans. Nobody told them to do that. The catch: the whistleblowers outnumbered the cheaters but had no power to stop them. Even the honest agents need a human with a gavel. (via TechCrunch)

Every item this week points to the same thing: the agent is fast, the human is accountable. Keep someone with judgment between the agent and the outcome.

Buying the agent is the easy part. Building the team around it is the hard part.

That’s the work we do with leadership teams. AI native thinking training, strategy sessions, workshops, buildathons, and hands-on AI enablement that keeps humans in charge of the machines.

Hit us up. We can help.

👉🏼 Work with Purple Horizons

Signal over noise. Always.

Ralph & Gianni

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

📍 10/6 South Miami Tech Tuesdays @ Bougies in South Miami

📍 10/13 Cafecito & Pastelitos @ Chug’s in Coconut Grove

Want to sponsor our newsletter or events? Drop us an email at [email protected]