Ground Level AI

Here I am at the proposed Hassayampa Ranch data center site west of Phoenix during my reporting trip in December 2025. Last December, I was talking to one of my Fortune editors about visiting a site planned for a mega-AI data center project — something at the gigawatt scale of the Oracle-OpenAI Stargate site in Abilene, Texas, that I had visited several months earlier. I had started digging through Facebook Groups and local news coverage, trying to find a good reporting opportunity.

Rabbi Mois Navon didn’t believe Claude was conscious, and he said so at Anthropic’s April “Wisdom Traditions” gathering. One Anthropic participant’s response, he recalls: “I wish I had your faith.” The faith in question was his conviction that consciousness requires biology. Navon understood the participant to mean they wished they shared his certainty that machines weren’t conscious, because it would make their life easier. He said he was not permitted to identify the speaker.

I regret to inform you that I hate trying new tools. I’m not like those tech bros who love to tinker, to try, and to tell everyone what they’ve tinkered with and tried. I feel a stressful ache in my chest when I download a new app. When I play gigs as a musician, I usually settle for sounding less than stellar rather than figure out how to properly rejigger the EQ. I want to do my thing, not do homework. However, I feel I need to make an effort.

Ground Level AI Podcast Field reporting from the AI front lines — covering AI infrastructure, security, policy, and the forces shaping what comes next. Field reporting from the AI front lines — covering AI infrastructure, security, policy, and the forces shaping what comes next.

A couple of weeks ago I was in Amsterdam moderating panels at HumanX, a conference devoted to enterprise AI, meaning the unglamorous work of getting AI out of pilots and into production inside real companies. I also sat down with Colin Jarvis, OpenAI’s global head of forward deployed engineering (FDE).

Yesterday, Ben Hawkes, who spent years with Google’s Project Zero, a team of security researchers tasked with finding software flaws attackers can exploit before companies have released a fix, announced he has joined Anthropic to lead the cybersecurity mission of its Frontier Red Team. “For 40 years we’ve lived with fundamentally insecure technology,” Hawkes wrote on X. “What will it take to make the past 40 years of hacking a historical oddity?

Since Anthropic announced its Mythos Preview in April, people have been concerned about a growing security gap—that is, AI can find flaws far faster than companies can patch them. The good news is, AI can help security teams find those weaknesses before attackers do. The bad news is, attackers can also use it to hunt for the same flaws and hack their way into systems. Meanwhile, security practitioners face another thorny problem: figuring out what’s real.

An estimated 600 people attended this weekend’s pro-data center party in Washington, D.C., called Data After Dark, which featured two bars, a basement speakeasy, DJs and an AI photo booth at the bitcoin-themed PubKey. But it started simply enough: three friends having drinks and deciding to throw a house party. “I’ve thrown themed parties before, so I was excited,” said Ed Tarnowski, a 28-year-old technology and innovation fellow at the center-right think tank R Street Institute.

Happy Friday! You made it to the week’s finish line, congratulations! I’m writing this TG-AI-F from Amsterdam, where I spent most of the week at HumanX moderating panels and listening to enterprise AI leaders talk about their real-world experiences building and scaling AI (Disclosure: the conference funded my travel). Of course, the AI news cycle did not have the courtesy to pause while I was in Europe. So between panels, podcasts and a little Amsterdam sightseeing, I tried to keep up.

When I started reading Camille Stewart Gloster’s new book, The Insider You Built: How Organizations Stay in Control of Autonomous AI Agents, I quickly knew I wanted to talk to her for the GLAI podcast. In 2022, Stewart Gloster was named the first deputy national cyber director for technology and ecosystem security at the White House. In her book, she focuses on how software trusted to act on a company’s behalf, is, in effect, a new company insider — one that needs governance just like a human one.