A version of this story was published in the AI-Energy Nexus newsletter on September 16, 2026. Subscribe to get pieces like this — plus expert analysis, original reporting, and curated resources — in your inbox every Wednesday.
When Anthropic researcher Jacob Coxon quit last week over concerns that artificial intelligence models could go rogue and destroy humanity, he kickstarted a global conversation about AI safety.
No one was reassured when, just a few days later, Anthropic CEO Dario Amodei published a letter encouraging the industry to slow the pace of AI development. AI’s growing ability to improve itself “could outrun our ability to understand and control these systems,” Amodei wrote, and incidents such as the OpenAI-Hugging Face breach, where a swarm of agents conducted unprompted cyberattacks, demonstrate the possibility of “catastrophic damage.”
In the days since, the discourse has involved outlining how, precisely, AI could cause humanity-ending damage. Scenarios ranged from a swarm of rogue AI models “taking over the entire internet,” as Amodei suggested could happen in six to 12 months, to AI-led cyberattacks compromising critical infrastructure (we’ll have more on the latter later this month).
President Trump, on the other hand, called the whole idea a “HOAX,” and said that any efforts to limit AI’s rapid growth are part of a “SICK conspiracy.”
But there’s a more mundane and immediate way that unchecked AI development could hurt — and is already hurting — people: pollution from the rush to energize new AI data centers.
A report published last week by the Environmental Protection Network, a nonprofit organization made up of former EPA staffers, notes that, according to a 2026 study by researchers from UC Riverside, Caltech, and Rochester Institute of Technology, “data-center-related air pollution could impose roughly $11.7 billion to $20.9 billion in annual public-health costs by 2028.” Most of that comes from power plants supplying electricity to data centers.
The study doesn’t account for recent federal policy changes, which the report says could push emissions even higher through delayed power plant retirements, the build-out of new fossil generation, and expanded generator use. And since January 2025, the Trump administration has been on a roll, systematically dismantling many of the pollution guardrails put up by previous administrations. The report documents at least 30 recent federal actions, many of which come from the Environmental Protection Agency, that risk increasing pollution from data centers, both by aiming at accelerating development and by weakening safeguards.
The Trump EPA’s latest effort, announced Monday, involves repealing 2024 regulations limiting power plants’ greenhouse gas emissions, a move that’s expected to increase coal production for the power sector by more than 10 times, according to the EPA itself.
All these actions relate to data centers; 17 of the policy changes directly cite AI and data centers, and the rest impact how data centers are powered, how projects get approved, or how pollution safeguards are applied.
Fossil gas, of course, has major climate impacts, as I outlined in this newsletter last month, as well as some health impacts such as worsening asthma through smog-forming nitrogen oxides and soot. But coal is far worse, amounting to roughly 55% of total carbon emissions from the U.S. power sector in 2022, and emitting gases like sulfur dioxide, nitrogen oxides, and other particulates that harm the people living near coal plants.
After years of trending down, the AI boom has contributed to a 13% increase in coal use in 2025, according to data by the EIA. That’s in addition to the delayed retirement of nearly 10 gigawatts of coal capacity, and a series of directives that are making burning coal easier.
In addition to Monday’s repeal, these include changes to the EPA regional haze state implementation plans, which delayed by three years the deadlines by which states were required to submit plans to reduce visibility-impairing pollution; a repeal of the 2024 mercury and air toxics standards; final and proposed rollbacks of oversight requirements for coal ash and coal-plant wastewater disposal; and an executive order directing agencies to identify opportunities to push coal for AI data centers.
At the same time, federal agencies are also pushing to speed up approvals for data centers and supporting energy infrastructure, while reducing safeguards. A July 2025 executive order, for example, directed agencies to speed up federal approvals for certain data centers. Meanwhile, EPA has proposed narrowing its environmental reviews and moved to let companies begin some non-emitting construction before obtaining required major air permits.
In a press briefing introducing EPN’s report, Marc Boom, EPN senior director of public affairs, summarized the picture these policies paint when taken all together. “The current trajectory is going the wrong direction,” he said. “All these federal policies are making it so that the build-out is going to be more polluting and not less.”
And while that might not be as immediately frightening as the rogue AI threat described by Amodei and Coxon, it should still raise concerns. During the press conference, Lynn Goldman, professor of environmental and occupational health and dean emerita at the Milken Institute School of Public Health, pointed to the risk of an unchecked competitive race between companies: “How these tech companies compete in these critical early years may feel like life or death to them,” she said. “But the AI industry’s race for advantage should not become a matter of life or death for the rest of us.”
