The White House launched a new strategy Monday to protect American privacy — and critical infrastructure like the electric grid — from artificial intelligence gone wrong.
But as AI creeps into the energy sector, experts say much more is needed, including passage of new legislation on Capitol Hill, to safeguard Americans from data breaches made possible by the technology.
That’s because AI energy products rely on extensive consumer data, such as when homeowners typically turn the lights on and what products are being charged, to manage and deliver electricity most efficiently — without a specific federal privacy law to restrict access to sensitive data.
President Joe Biden’s executive order directs the Department of Energy, alongside the Department of Homeland Security, to “address AI systems’ threats to critical infrastructure,” while also funding research into privacy protection techniques. The announcement arrives as a growing chorus of experts, including staunch AI supporters in the energy sector, are warning that the technology could put sensitive data at risk for millions of Americans.
“We cannot take AI lightly at all,” said Scott Koehler, vice president of global strategy for the electricity grid at Schneider Electric, a power-sector behemoth that produces AI software.
“You just see how fast AI technology is maturing,” he said in an interview, urging new federal regulations to protect privacy. “It won’t benefit anyone if AI gets to the point of really doing serious damage.”
As products like ChatGPT — a free language processing tool that produces humanlike responses to random questions — grip the United States, Koehler and energy industry leaders are voicing concerns about AI, while at the same time proselytizing its potential to streamline electricity distribution and cut emissions.
The energy sector privacy risks are abundant.
The electric grid is already uniquely vulnerable to data breaches, and such intrusions on AI software tied to the grid — malign or otherwise — could lead to the unwanted disclosure of sensitive personal data like electricity billing records, home addresses and phone numbers — and energy use patterns that reflect when a person is home. All of that information could be used for phishing attempts or physical break-ins. At oil and gas operations where AI is also being deployed, breaches could reveal confidential business information and grind electricity production to a halt.
The AI debate is a delicate one. The technology isn’t just one thing: It describes a sweeping set of products that learn data and generate novel content, in some cases even refining and improving results over time. It’s already used across the health care and entertainment sectors.
With energy, there are a range of applications. AI software can be used to optimize electricity distribution based on supply and demand trends. At oil and gas operations, similar software can be used to sift through reams of data to identify and rectify a disruption. AI-powered drones can quickly assess whether damaged power lines should be replaced after storms, and the technology can analyze seismic data to determine optimal spots for oil and gas drilling.
The rapid emergence of AI also comes amid a regulatory vacuum that the Biden administration is seeking to fill.
“This is really a microcosm of what is possible in AI. We obviously have to get the security of AI in critical infrastructure right. We are going to do that,” a senior Biden administration official told reporters Sunday. “There’s a lot of opportunity here as well. And AI in critical infrastructure and in the power grid in particular, I think, shows that.”
The official, who wasn’t authorized to speak by name about the White House’s plans, mentioned AI systems run by Alphabet Inc. that, in past years, slashed energy use at data centers. Alphabet is the parent company of Google.
The executive order also forces companies that develop “the most powerful AI systems” to share results with the government, along with a host of other directives for agencies to spur innovation, hire AI experts, assert U.S. leadership abroad and protect privacy. The official called on Congress to pass privacy legislation.
“Without safeguards, AI can make it easier to extract, identify and exploit personal data, and it can heighten incentives to do so,” the official said.
Helena Fu, director of critical and emerging technology at DOE, recently said the department plans to announce new AI policies alongside the White House executive order. DOE didn’t provide a statement to E&E News on Sunday.
Privacy experts say authorities available to a president are limited, arguing that new legislation on Capitol Hill is necessary to protect Americans from AI-caused data breaches. The privacy risk in the energy sector comes from potentially faulty AI products that disseminate data, including from Chinese malware projects that use AI, according to experts.
“There are risks especially within the energy space,” said Brandon Pugh, policy director for cybersecurity and emerging threats at the R Street Institute, a free-market think tank. “There’s definitely a different risk for critical infrastructure having a disruption than potentially a local business. Both are serious, but there’s definitely a distinction.”
At a congressional hearing this month on AI and the energy sector, Paul Dabbar, DOE’s undersecretary for science during the Trump administration, warned that “China can place backdoor physical ability and chips to break into systems that they shipped to us.”
The emissions factor
In the backdrop of the AI rise, climate change looms. Biden aims to achieve 100 percent carbon-free U.S. electricity by 2035 as part of an effort to avert catastrophic climate change.
That means mass electrification aimed at displacing natural gas heating and gasoline-powered vehicles with electric vehicles and renewable energy like solar. It also means the personal data for Americans, like the times people are home to turn on the lights and appliances, could be more vulnerable to theft or other misuse.
The federal government, in many ways, is racing to catch up with the private sector on AI. Ahead of Australian Prime Minister Anthony Albanese’s visit to the United States last week, the White House pledged to work with Australia to advance AI development for pandemic preparedness, drought resilience and other issues.
Earlier this month, DOE doled out a $250 million grant to improve electricity distribution in Oregon with AI. The department is also pursuing dozens of AI research initiatives, from electricity grid cybersecurity protections to methane leak detection and carbon-dioxide-induced enhanced oil recovery.
While Koehler concedes the potential perils, he — and many others in the power sector — say AI needs to be embraced.
“It’s becoming an ever-increasing area of investment in innovation,” he said. “As more data is collected and more history of information is available, you’re continuing to refine, improve and start to automate how things improve, meaning how you predict where energy will be and the fluctuations.”
Others note various ways the technology could bolster renewable and low-carbon energy. At the hearing, Dabbar said AI could improve wind operations significantly by allowing for monitoring of “every gear, every turbine blade, hour by hour,” for example. Similarly, adding AI sensors at the turbines of nuclear power plants could make it much easier to detect when a part is aging or failing — lowering maintenance costs significantly in the power sector, he said.
A report last year from McKinsey & Co., a consultancy juggernaut for government agencies and businesses alike, concluded that AI software deployed at the Martin Lake power plant in Texas abated hundreds of thousands of tons of CO2 emissions. A Vistra subsidiary operated the technology, which optimizes the thermal efficiency of the plant.
Experts say AI could also play a vital role in energy systems like virtual power plants, which aggregate and manage distributed energy resources like solar panels, EV chargers and smart appliances, and microgrids, which do the same without a connection to the grid. Both systems can use renewable energy to manage and distribute power in a more optimized way, thus decreasing energy use and relieving the grid.
Schneider produces a range of software and electrical panels for such systems, including tools for residential electricity, data center electricity, grid connection and other purposes.
Many other companies are involved, too. Utilidata, a company based in Providence, R.I., produces a microchip that it says will improve efficiency of electricity distribution with demand forecasting and other tools, all of which can save big amounts of money.
“This is a perfect place, in our view, for DOE to use some of the funding that they have to accelerate this kind of technology,” said Jess Melanson, the chief operating officer at Utilidata. “It is newer to the grid, but it’s not new to other industries. And the grid desperately needs it to get ready for this massive change that’s happening all around it.”
Utilidata is angling for a piece of the yet-to-disbursed $3 billion in “smart grid grants” enacted in the 2021 bipartisan infrastructure law.
Meanwhile, the Canadian data firm Shinydocs is deploying AI to manage and retrieve engineering and safety data that’s necessary in the oil and gas business. It is selling software to U.S. companies like Southern California Gas and Valero Energy to streamline data management traditionally done by humans, according to Shinydocs CEO Jason Cassidy.
“It’s always been done with just humans tracking down information manually, and then verifying it,” Cassidy said. “Now that a lot of information has been digitized, it opens the way for training a machine to have that same level of human intelligence.”
Cassidy says his software purges sensitive data “with certainty.”
‘No comprehensive privacy law’
So far, AI is delivering mixed results for decarbonization, said Jeremy Renshaw, a senior technical executive for AI at Electric Power Research Institute, a nonprofit research and development group.
“AI is not something that has only benefits and no drawbacks,” said Renshaw. “There are many benefits for using AI to help with energy efficiency, with things like demand response and virtual power plants.”
Renshaw called workforce training a major priority in preventing misuse of data with AI products.
“We know how to implement and use data. We know how to anonymize data, and we know how to train and test and validate so that the model is not either making connections that would divulge information or that is not divulging information that it shouldn’t,” he said.
But not only are major AI regulations nonexistent, they’re unlikely to come anytime soon, regardless of Biden’s executive order.
“Outside of specific sectors and then state-level legislation, there’s no comprehensive privacy law in the United States, meaning that there’s no law governing how people’s data is secured [and] allowing them to weigh in on terms of their consent,” said R Street’s Pugh. “[Places] around the world like China and the European Union … are well beyond [where] we are.
“From a consumer perspective, that’s not ideal,” he said.
Senate Majority Leader Chuck Schumer (D-N.Y.) is now leading a legislative effort on AI. In June, he laid out some guidelines for bills.
“The rapid pace at which AI is advancing presents unique challenges,” Schumer’s office said at the time. “With so much potential, the U.S. must lead in innovation and write the rules of the road on AI and not let adversaries like the Chinese Communist Party craft the standards for a technology set to become as transformative as electricity.”
There’s some bipartisan work to build on. Last year, House Energy and Commerce Committee lawmakers teamed up to pass landmark privacy legislation, the “American Data Privacy and Protection Act,” out of committee. While the bill didn’t address AI specifically, it could potentially mitigate some of the privacy threats posed by the technology.
Now the committee — back in full operation after a three-week Republican battle to seat a speaker — is hosting a series of hearings to flesh out congressional direction on AI. E&C Chair Cathy McMorris Rodgers (R-Wash.) echoed Schumer’s concerns about Chinese technology.
“We will explore why a national data privacy standard is foundational to both protecting people’s data privacy and promoting innovation,” Rodgers said last week in a statement. “It’s how America — not China — leads the future on artificial intelligence.”
To kick off the hearing in mid-October where Dabbar testified, Rep. Jeff Duncan (R-S.C.), chair of the Energy, Climate, and Grid Security Subcommittee, delivered a plea — and an admission that Congress is ill-prepared to tackle AI.
“The committee will get more benefit from hearing from you than hearing ourselves talk,” Duncan told the witnesses. “We really don’t know a lot about this subject.”