Reporting requirements are necessary to alert the government to potentially unsafe recent capabilities of increasingly powerful artificial intelligence models, says a U.S. government official who works on artificial intelligence issues. The official, who asked for anonymity and the ability to speak freely, points out Introduction to OpenAI about the “inconsistent refusal to submit applications for the synthesis of nerve agents” in its latest model.
The official claims that the reporting obligation is not excessively burdensome. They argue that unlike AI regulations in the European Union and China, Biden’s EO reflects a “very broad, light approach that continues to support innovation.”
Nick Reese, who served as the Department of Homeland Security’s first director of emerging technologies from 2019 to 2023, rejects conservative claims that mandatory reporting would threaten companies’ intellectual property. He also says it could actually benefit startups by encouraging them to develop “more computationally efficient” and less data-heavy AI models that fall outside the reporting threshold.
The power of artificial intelligence makes government oversight indispensable, says Ami Fields-Meyer, who helped draft Biden’s EO as a White House technology official.
“We’re talking about companies that claim to be building the most powerful systems in the history of the world,” Fields-Meyer says. “The first duty of government is to protect the people. “Trust me, we’ve got this” isn’t a particularly compelling argument.
Experts praise NIST security guidelines as an important resource for building security in new technology. They note that faulty AI models can cause serious social harm, including rental and lending discrimination and unjustified loss of government benefits.
Trump’s own first-term order on artificial intelligence required federal AI systems to respect civil rights, which will require research on social harms.
The AI industry has largely welcomed Biden’s security agenda. “We hear that spelling these things out is generally useful,” says a U.S. official. For recent companies with miniature teams, “it increases the ability of their employees to solve these problems.”
Withdrawing Biden’s EO would be a troubling sign that “the U.S. government is going to abandon its approach to AI security,” says Michael Daniel, a former presidential cybersecurity adviser who now heads the Cyber Threat Alliance, a nonprofit information sharing organization.
When it comes to competition with China, EO defenders say the security rules will actually support America win by ensuring that U.S. artificial intelligence models perform better than their Chinese rivals and are protected from Beijing’s economic espionage.
Two very different paths
If Trump wins the White House next month, expect a radical shift in the government’s approach to AI security.
Republicans want to prevent AI-related harms by applying “existing tort and statutory laws” rather than introducing sweeping recent restrictions on the technology, Helberg says, and favor “a much greater focus on maximizing the opportunities that AI brings, instead of focusing too much on risk.” mitigation.” This would likely spell doom for the reporting requirement and perhaps some NIST guidelines.
The reporting requirement could also face legal challenges now that the Supreme Court has eroded the deference that courts have shown agencies in evaluating their regulations.
And GOP opposition could even threaten NIST voluntary partnerships in artificial intelligence testing with leading companies. “What will happen to these obligations in the new administration?” – asks an American official.
This polarization around artificial intelligence is frustrating technologists who fear Trump will undermine the search for safer models.
“There are risks associated with caring for artificial intelligence,” says Nicol Turner Lee, director of the Center for Technology Innovation at the Brookings Institution, “and it is critical that the next president continues to ensure these systems are safe and secure.”