The Big Story: Policymakers parse liability for AI developers
Policymakers are putting forward AI liability proposals that could impact how models are developed by top labs, shared across the ecosystem, and deployed by startups. Liability-based proposals are gaining steam as a response to recent cyber incidents involving AI agents, but the proposals merit careful consideration of how they will impact the various actors across the AI ecosystem. In a new blog post this week, Engine highlights principles that can guide stakeholders as they examine the proposals—such as accounting for the various actors in the ecosystem, balancing incentives, accounting for both risks and benefits, and considering policy alternatives.
House and Senate lawmakers each have proposals out this month designed to hold developers liable when their AI systems cause harm, but to varying extents acknowledge there may be instances where others or users are responsible. Some have suggested creating strict liability for AI—holding developers liable regardless of fault. Others—including the top Democrat on the Senate Commerce Committee and leaders of the administration's newly formed “SI Force”—have suggested that existing laws apply to AI companies.
These proposals may be conceived with the largest companies in mind, but there are several entities in the AI value chain that must be considered. Liability should be tailored to the actors most responsible for and capable of preventing harm. Liability changes incentives, and frameworks should incent developers to test rigorously and address foreseeable risks without discouraging model development, sharing, or beneficial uses of the technology. Moreover, assigning liability is only one potential tool policymakers have, and they should also consider other approaches to shore up cybersecurity in a world where AI capabilities are only advancing. Congress is out until after the election—and it is unlikely any major proposals will advance when they return—but the debate over AI liability will continue, making it important that policymakers account for the startups building and deploying these technologies.
Policy Roundup:
More importers could recover unlawfully collected tariffs. Businesses that paid tariffs struck down by the Supreme Court could have a new path to refunds after a federal trade court certified a class covering importers who have not filed individual lawsuits. The decision could be particularly important for smaller importers, including startups, as the judge warned that some may not pursue refunds through the existing process because of the cost, effort, or lack of awareness involved. Some $30 billion in unlawfully collected tariffs is still held by the government.
Agency suspends tech companies from green card program. The U.S. Department of Labor announced it will not accept new or process pending permanent labor certification applications from several companies including Microsoft and Adobe. The certification is the first step for most employer-sponsored green cards, and the suspensions have no set end date. They come amid broader changes to employment-based immigration that add cost and uncertainty to pathways startups already struggle to use to access international talent.
Appeals court allows Georgia, Florida age verification laws to be enforced. In separate challenges this week, a federal appeals court overturned preliminary injunctions against two state laws requiring Internet platforms with certain features and content to verify the age of their users. Although the laws were likely written with the largest social media platforms in mind, they apply to services that have certain features, such as the ability to create a user profile or showing “like” counts on posts. To comply, in-scope services must verify the ages of all of their users, a requirement that for startups strains budgets, heightens cybersecurity risks, and undermines competitiveness.
Billions in resources announced to support AI research. Technology companies, government agencies, and philanthropic leaders pledged billions in computing resources, AI model access, cloud credits, and training programs to support AI innovation and research. Expanding access to these resources will help accelerate scientific research, leading to commercializable innovation and growing the nation's AI talent pool.
Restrictions for young users fracture Internet. Turkey enacted a law banning children under 15 from social media beginning November 1, requiring platforms to implement age verification measures—including through a government ‘token’—to enforce the restrictions. As more countries adopt different rules for young users, the growing patchwork risks fracturing the Internet, and startups could face mounting costs to verify users' ages and adapt their services to comply with each country's requirements, potentially limiting where they can afford to operate.
Startup Roundup:
#StartupsEverywhere: Coral Gables, Florida. DraftyAI builds research and drafting software for immigration attorneys, turning their evidence and case information into complete legal briefs, including exhibit lists. We sat down with co-founder and CEO Nadine Navarro to discuss her path from running a solo immigration practice to building legal tech, the AI and data privacy questions facing startups that handle sensitive client information, and what U.S. immigration policy means for startups, as well as international talent.