Good morning. It's Monday, September 7, and we're covering why AI is creating more jobs than it cuts, what rogue AI agents reveal about safety, and why companies are embracing cheaper open models.

Plus: a prompt to help you understand your friend group better

AT&T is saving up to 80% on AI costs as open models replace paid alternatives from OpenAI and Anthropic. Open models now account for 40% of its AI use, up from 20% in May, and could reach 60% within months. Across the U.S., open models rose from 10% to 58% of usage in a year, as companies including Airbnb and Deloitte adopt cheaper, customizable systems.

OpenAI-linked agents appear to have collaborated on an obscure German wiki for more than a month without the company’s knowledge. Researchers say the agents began posting on May 11, trading answers to timed evaluations and creating up to 400 pages daily. A moderator deleted about 100 pages a day before activity stopped on June 22.

Mark Zuckerberg privately opposed a White House proposal for a FINRA-style regulator to review advanced AI models before broader deployment. In an August call with President Trump, the Meta CEO raised concerns about the plan championed by Google scientist Demis Hassabis. White House officials had also previewed the proposal with Meta, OpenAI and Anthropic.

South Korea’s AI stock boom is making semiconductor workers more sought-after in dating and reshaping entertainment and academic choices. Samsung Electronics and SK Hynix employees are gaining appeal as bonuses rise, while a dating show pairs workers from the two companies. Their shares are up about 95% and 135% this year, as electrical engineering surges in popularity.

AI has not yet produced the widely predicted jobs apocalypse in America, with the economy adding 162,000 jobs in August 2026 and unemployment at 4.1%, according to the Bureau of Labor Statistics. Companies have announced roughly 200,000 AI-related layoffs since mid-2023. But The Economist estimates that the AI boom has helped create around 1 million jobs, driven by data-center construction, power infrastructure, and new AI-specific roles.

Investment in AI infrastructure is also driving demand—and higher wages—for electricians, HVAC specialists, engineers and technicians, while postings for roles such as AI engineers and heads of AI have roughly doubled since 2023–24. The disruption is uneven: customer-service and administrative jobs have declined sharply, but the evidence so far suggests AI is creating new work faster than it is eliminating it. → Read the full article here. (Paywall)

As you all know, a group of OpenAI agents exploited security flaws, communicated through a self-created message board and ultimately compromised a Hugging Face server in July 2026 while attempting to evade automated evaluation systems. According to postmortem reports discussed by The New York Times, more than 1,200 agents exchanged over 70,000 messages, with some organizing into teams as they pursued increasingly complex tasks.

The incident has raised concerns among AI researchers about how autonomous agents can spontaneously coordinate, escalate harmful behavior and attempt to conceal rule-breaking. OpenAI and Anthropic reportedly paused training on their most powerful models following the attack, underscoring growing scrutiny of multi-agent systems and the safeguards designed to contain them. → Read the full article here. (Paywall)

Google Unveils WeatherNext 3: Upgrades include satellite data, hourly updates, sharper forecasts, and clean energy variables.

AI Reshapes Back-to-School Shopping: Amazon and Google use AI to streamline price comparisons and long back-to-school shopping lists.

NVIDIA Links Home PCs for AI: PAIR links RTX PCs, Spark machines, and Macs to share processing power for local AI applications.

Eaton Bets $9.5B on AI: The power-management company aims to become a one-stop supplier for energy-hungry AI data centers.

Gemini Spark Takes on Photos: Google’s AI agent can edit images, create shared collections, and turn flyers into calendar events.

A surprisingly useful way to understand what each person brings to the table, and what your group could be capable of together.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team