AI Policy Weekly #19
Policy framework to mitigate extreme AI risks, $6.4B in CHIPS funding to Samsung, and a new bill to strengthen export control enforcement
Welcome to AI Policy Weekly, a newsletter from the Center for AI Policy. Each issue explores three important developments in AI, curated specifically for AI policy professionals.
Senators Introduce Framework for Mitigating Extreme AI Risks
“Any comprehensive framework to address risks from AI should also include measures to guard against the potential catastrophic risks,” reads a new letter from Senators Jerry Moran (R-KS), Jack Reed (D-RI), Angus King (I-ME), and Mitt Romney (R-UT).
The Senators are specifically concerned about upcoming AI systems facilitating the development of biological, chemical, cyber, and nuclear weapons.
To illustrate their concern, they point to reports and remarks from the Department of Defense, State Department, US Intelligence Community, Department of Energy, Department of Homeland Security, United Kingdom, White House AI Executive Order, and elsewhere, highlighting how terrorists and adversarial nations could weaponize cutting-edge AI systems to pose serious national security threats in the future.
This bipartisan Senate quartet also offers a concrete policy proposal to address these emerging threats.
Their first recommendation targets AI hardware. They propose that entities selling or renting massive quantities of computational power must report their activities and “exercise due diligence to ensure that customers are known and vetted.” The Commerce Department is already working to establish a scheme like this.
Their next recommendation targets the creation of extremely computationally intensive AI systems. Developers of such systems must notify the government of their plans, incorporate safeguards against weaponization, and implement cybersecurity to prevent theft.
Finally, the Senators propose that massive AI models should require government approval before deployment. A new regulatory entity should license major AI models, ensuring that their deployment is contingent upon effective safeguards against biological, chemical, nuclear, and cyber threats.
The oversight entity could be a new interagency coordinating body, akin to the Committee on Foreign Investment in the US (CFIUS). Or it could be a preexisting federal agency like the Department of Commerce or the Department of Energy. Alternatively, “since frontier models pose novel risks that do not fit neatly within existing agency jurisdictions,” Congress could establish a new agency for this purpose.
This isn’t the first time Congress has seen a proposal for an independent oversight body to license advanced general-purpose AI models: Senators Richard Blumenthal (D-CT) and Josh Hawley (R-MO) called for this in their own AI policy framework last September.
However, the Blumenthal-Hawley framework was broader, with additional recommendations regarding legal liability, export controls, information disclosure, watermarking, and more. It targeted a wider variety of AI risks, whereas the Romney-Reed-Moran-King proposal focuses primarily on catastrophic risk.
The Center for AI Policy supports both of these frameworks—which share priorities with our own model legislation for AI—and urges Congress to prioritize passing their components into law.
Samsung Receives Billions in CHIPS Funding
Now that another week has passed, dedicated AI Policy Weekly readers will be unsurprised to hear that billions of additional dollars have been injected into America’s AI future.
This week’s lucky winner is Samsung, which signed a non-binding deal to receive $6.4 billion in CHIPS Act funding.
Perhaps the luckiest winner is a tiny Texas town named Taylor, where Samsung will expand its empire.
“It’s a little city of manufacturing,” boasted Commerce Secretary Gina Raimondo.
In Taylor, Samsung will build two leading-edge chip factories, one advanced packaging facility, and one R&D factory dedicated to plotting the future of chips.
The company announced plans in 2021 to build in Taylor, and the boost of $6.4 billion plus a 25% investment tax credit will ensure that the project achieves its full potential.
In total, the Commerce Department has spread $21.5 billion of CHIPS funding across TSMC, Intel, and Samsung to build cutting-edge facilities. These three semiconductor giants are the world's only companies capable of manufacturing leading-edge chips at scale.
The CHIPS funding marks an effort to lift America from manufacturing 0% to manufacturing 20% of the world’s most advanced chips by 2030.
Romney, Hassan Seek to Fortify Export Controls on AI and Chips
13.5 years ago, then-President Barack Obama issued Executive Order 13558, which established the Export Enforcement Coordination Center (E2C2) within the Department of Homeland Security to do exactly what its name suggests: harmonize export control enforcement efforts across the federal government.
The E2C2 opened in 2012 and continues to operate today, but it was never formally established.
To solidify its role, Senators Mitt Romney (R-UT) and Maggie Hassan (D-NH) introduced a new bill that would formally establish the E2C2, authorizing $25 million in FY25 appropriations.
Importantly, the bill would also assign the Center specific responsibilities to curb the evasion of export controls, particularly targeting the smuggling of semiconductors, advanced AI, and quantum technology to US adversaries like China, Russia, Iran, and North Korea.
Given the growing geopolitical importance of AI and chips, this project could be critical for US national security.
News at CAIP
From 11am–12pm on Tuesday, April 23rd, we will host a moderated discussion on AI, Automation, and the Workforce in SVC 212 inside the Capitol Visitor Center. The speakers will be Professor Simon Johnson of MIT and Professor Robin Hanson of GMU. To attend, fill out an RSVP using this form.
We’re hiring for two different roles: External Affairs Director and Government Relations Director.
Our latest blog post explores the American public’s support for AI regulation.
Jakub Kraus coauthored a review article on AI governance in the Annual Review of Political Science.
Jason Green-Lowe spoke at a panel discussion examining AI’s impacts on social security, as part of the Social Security Administration’s National Disability Forum.
Quote of the Week
The really disorienting thing about talking to the people building AI is their altered sense of time. You’re sitting there discussing some world that feels like weird sci-fi to even talk about, and then you ask, well, when do you think this is gonna happen? And they say, I don’t know—two years.
—Ezra Klein, New York Times Opinion Columnist, prefacing his podcast interview with the CEO of Anthropic
This edition was authored by Jakub Kraus.
If you have feedback to share, a story to suggest, or wish to share music recommendations, please drop me a note at jakub@aipolicy.us.
—Jakub