Remember when everybody at the state Capitol was seemingly railing against tax incentives for data centers?
In her State of the State Address, Democratic Gov. Katie Hobbs announced she wanted to end the “$38 million dollar corporate handout” and praised Republicans for agreeing with her.
Republican Rep. Neal Carter introduced a bill to block any new incentives after the end of the year. Democratic lawmakers were close behind, introducing seven bills that took aim at those incentives.
It seemed like a rare bipartisan moment of agreement.
But lawmakers did not get to work on ending those tax credits.
None of those bills were heard in the Republican-controlled committees ahead of the deadline to hear them and keep them alive. Most of them died without ever getting a hearing in Republican Rep. Justin Wilmeth’s AI and Innovation Committee.
“Those bills would have damaged Arizona’s competitive standing in the marketplace,” Wilmeth told us. “Thirty-two other states offer data center tax incentives and repealing our incentives would not just hurt the industry, but countless others that data centers support.”
This week marks the first time since the legislative session started that the public is getting a clear view of which bills actually have a shot at becoming law.
Until now, it’s been a free-for-all as lawmakers introduced a record number of bills. But after this week, most of those bills won’t see the light of day again.
What’s happening right now is the House and Senate are passing a flurry of bills and sending them to the other chamber. If a bill doesn’t move into the other chamber this week, then it’s basically dead.
That means we can see which of the AI-related bills are still going strong, and which ones are being left on the cutting room floor.
Taking stock
The Legislature took a step forward this year by forming the House AI and Innovation Committee, which gave lawmakers a formal setting to discuss AI policies.
For the most part, they’ve taken a pragmatic approach to AI, as we wrote last week. There isn’t a lot of hyperbole at committee hearings about AI destroying, or saving, the world. They’re mostly looking at use-cases and trying to reduce harm.
Broadly put, lawmakers are signing off on bills that deal with the social issues that come with AI, like banning deepfakes and requiring public schools teach students how to use AI tools.
But they’re struggling with how to handle the infrastructure that powers AI, particularly data centers that are causing local controversies.
So far, here are the bills that the House already passed, and sent to the Senate for consideration:
Republican Rep. Nick Kupper’s HB2133 would ban pornographic deepfakes.
Republican Rep. Tony Rivero’s HB2311 would require AI companies to clearly tell kids that they are interacting with AI, among other safeguards.
Republican Rep. Alexander Kolodin’s HB2409 would create a summer school course through the Arizona Artificial Intelligence Education Program.
Kolodin’s HB2410 would protect your privacy when you ask an AI chatbot for legal advice.
Three other bills have already passed committee and likely will be passed by the full House this week.
Republican Rep. Teresa Martinez’s HB2371 would allow couples to use AI-assisted arbitration during divorce proceedings.
Republican Rep. Justin Wilmeth’s HB2592 would require state agencies to find ways to use AI to streamline administrative tasks and to find regulations that unnecessarily restrict AI innovation.
Kupper’s HB4005 would require public schools to offer instruction on the ethical, moral, and educational uses of AI.
Dealing with data centers
Most of the bills that have gained traction deal with relatively uncontroversial topics. Nobody wants to find out somebody made sleazy deepfake images of them, and people generally agree that students should get some version of training on how to use AI.
It’s the infrastructure issues that are really raising a ruckus.
Arizonans are clamoring for protection against skyrocketing electric bills and the draining of their aquifers as data centers — backed by multi-billion-dollar companies — pop up all over the state.
And lawmakers have proposed some ideas to deal with those issues.
It’s just that none of the bills that would rein in data centers, most of which were filed by Democrats, are still alive. The Republican chairs of the committees that oversee AI bills aren’t letting any of them through.
Instead, the bills that were up for a vote during a marathon session last night were designed to make it easier to build more data centers.
Republican Rep. Jeff Weninger’s HB4009 would require the State Land Department to develop a map that shows state trust land parcels that are suitable for data centers. State officials would make that map in consultation with the data center industry.
Wilmeth’s HB2452 would require Arizona’s most populous counties to update their comprehensive plans to include land for data centers.
Crossover week isn’t the end of the legislative session, of course. Now, Senate committees will hear bills from the House, while House committees hear bills from the Senate. The lucky bills that survive that process will head to the governor get signed or face her veto stamp.
Like always, you can follow along by using our bill tracking list, which we compiled with our legislative intelligence service, Skywolf.
Big, if true: President Donald Trump says he worked out a deal with major tech companies to pay more for energy costs as data centers pop up all over the country, per the New York Times. The president didn’t offer any details about the agreement, but it’s a signal that he is at least aware that communities where data centers get built are worried about seeing their electric bills go up.
Coming at it from a different angle: Democratic U.S. Sen. Ruben Gallego is pushing to make it easier to build nuclear power plants to deal with the AI boom’s energy needs, Greg Hahne reports for KJZZ. Gallego’s bill would cut down on the risk of delays or projects going over budget. That federal help could come in the form of federal funding, which would be music to the ears of the big utility companies in Arizona that announced last year they planned to build nuclear power plants.
Checking in: For the past two years, several assisted living facilities in the Phoenix area have used an AI-powered motion detector to help reduce the risk of falls, and the track record so far is promising, Jessica Boehm reports for Axios. Before the motion detectors were installed, one of the facilities in Mesa reported about 20 falls per month. Now, it’s down to about three per month.
I, teacher: Two Arizona schools are leaning hard on AI instructors, Arizona Family’s Derek Staahl reports. At Unbound Academy and Novatio School, students spend two hours a day learning from their AI teachers and school officials say their students are learning faster than those at other schools.
We like to think AI Agenda readers learn about AI faster than most people. If that rings true to you, why not help us out by upgrading to a paid subscription?
Writing, on the double!: At Fort Huachuca in southeastern Arizona, Army doctrine writers are using AI to write training guidance, often in half the the time they were used to, KGUN’s Alexis Ramanjulu reports. They’re using AI as a collaborative tool to “get my rough drafts out and to have someone to speak with when I'm first forming my ideas before they go on paper,” as one writer put it.
AI labs are suddenly discovering what copyright lawyers have been yelling about for three years.
Anthropic says it has detected large-scale “distillation attacks” targeting its Claude models — essentially, competitors hammering the system with millions of prompts to train their own models on the outputs.
Not normal usage. Not hobbyists. Industrial-scale “ask it everything” campaigns.
The AI arms race, apparently, now includes model counter-intelligence.
The irony is sitting right there in plain sight. Large language models were trained on vast stretches of the open internet — blog posts, forums, fan fiction, recipe sites, your 2012 Tumblr phase — and now the same industry is drawing hard property lines around its own generated text.
Scrape the web? Innovation. Scrape the model? Attack.