Center for Data Innovation Blog

The temporary shutdown of Anthropic's Fable 5 and Mythos 5 highlighted the uncertainty surrounding the government's authority to restrict access to frontier AI models and the costs that uncertainty can impose on innovation and trust. To provide greater clarity for developers, customers, and allies, Congress should establish a transparent statutory framework for AI shutdown authority with clear standards and proportionate safeguards.

Adversarial AI distillation poses a growing threat to U.S. AI leadership and national security, and Congress is right to address it through the Deterring American AI Model Theft Act (DAAMTA). But before passing the bill, lawmakers should narrow its scope, strengthen evidentiary standards, protect legitimate research, and pair it with technical safeguards and international cooperation to better combat large-scale AI model theft without hindering innovation.

As employers increasingly seek workers who can apply AI alongside domain expertise, universities should integrate AI across disciplines—not just standalone AI or computer science programs—to prepare graduates for the modern workforce.

Adversarial AI distillation poses a growing threat to U.S. technological leadership, national security, and AI safety by enabling foreign actors to extract the capabilities of frontier models without authorization. Policymakers should strengthen legal and technical defenses against industrial-scale model theft while ensuring any response preserves legitimate AI research, innovation, and the responsible use of distillation.

Cities that are successfully scaling AI are investing in workforce upskilling alongside governance and technology deployment. Case studies from Washington, DC, San Jose, Seattle, and Cleveland show that employee training and AI literacy are critical to turning pilot projects into lasting improvements in public service delivery.

Unlike training-data disputes, CNN's lawsuit against Perplexity alleges near-verbatim reproduction of its journalism through AI search products. Policymakers should favor targeted enforcement—not sweeping AI restrictions.

Five states—Utah, Connecticut, Ohio, Texas, and North Carolina—are showing how centralized AI sandboxes, oversight frameworks, and clear evaluation metrics can help governments move beyond isolated pilot programs and scale AI tools to deliver measurable improvements in public services.

As AI-powered cyber threats become more advanced, the federal government should modernize the CyberCorps SFS program by integrating AI-security training, reforming cyber hiring pipelines, and expanding training infrastructure to build a stronger cybersecurity workforce.

OECD data shows there is a consistent, positive relationship between the share of firms using AI and a country’s GDP per hour worked.

Requiring government approval before releasing advanced AI models would slow innovation, politicize AI development, and weaken U.S. competitiveness. Instead, policymakers should focus on collaborative safety efforts and strengthening cybersecurity.