Computer scientists have long sought to build systems that can actively and autonomously carry out complicated goals in the real world—commonly referred to as artificial intelligence "agents." Recently, significant progress in large language models has fueled new optimism about the prospect of building sophisticated AI agents. This CSET-led workshop report synthesizes findings from a May 2024 workshop on this topic, including what constitutes an AI agent, how the technology is improving, what risks agents exacerbate, and intervention points that could help.

As critical infrastructure operators and providers seek to harness the benefits of new artificial intelligence capabilities, they must also manage associated risks from both AI-enabled cyber threats and potential vulnerabilities in deployed AI systems. In June 2024, CSET led a workshop to assess these issues. This report synthesizes our findings, drawing on lessons from cybersecurity and insights from critical infrastructure sectors to identify challenges and potential risk mitigations associated with AI adoption.

In February 2024, CSET introduced its new cybersecurity jobs dataset, a novel resource comprising ~1.4 million LinkedIn profiles of current U.S. cybersecurity workers. This data snapshot uses the dataset to identify top-producing institutions of cybersecurity talent.

President Biden's October 2023 executive order prioritizes the governance of artificial intelligence in the federal government, prompting the urgent creation of AI risk management standards and procurement guidelines. Soon after the order's signing, the Office of Management and Budget issued guidance for federal departments and agencies, including minimum risk standards for AI in federal contracts. Similar to cybersecurity, procurement rules will be used to enforce AI development best practices for federal suppliers. This report offers recommendations for implementing AI risk management procurement rules.

China’s brain-computer interface research has two dimensions. Besides its usual applications in neuropathology, China is extending the benefits of BCI to the general population, aiming at enhanced cognition and a “merger” of natural and artificial intelligence. This report, authored in collaboration with researchers from the Department of War Studies at King’s College London uses bibliometric analysis and expert assessment of technical documents to evaluate China’s BCI, and conclude that the research is on track to achieve its targets.

Research participants who read propaganda generated by GPT-3 davinci (a large language model) were nearly as persuaded as those who read real propaganda from Iran or Russia, according to a new peer-reviewed study by Josh A. Goldstein and co-authors.

This data snapshot is the first in a series on CSET’s cybersecurity jobs data, a new dataset created by classifying data from 513 million LinkedIn user profiles. Here, we offer an overview of its creation and explore some use cases for analysis.

Strengthening the federal cyber workforce is one of the main priorities of the National Cyber Workforce and Education Strategy. The National Science Foundation’s CyberCorps Scholarship-for-Service program is a direct cyber talent pipeline into the federal workforce. As the program tries to satisfy increasing needs for cyber talent, it is apparent that some form of program expansion is needed. This policy brief summarizes trends from participating institutions to understand how the program might expand and illustrates a potential future artificial intelligence (AI) federal scholarship-for-service program.

While recent progress in artificial intelligence (AI) has relied primarily on increasing the size and scale of the models and computing budgets for training, we ask if those trends will continue. Financial incentives are against scaling, and there can be diminishing returns to further investment. These effects may already be slowing growth among the very largest models. Future progress in AI may rely more on ideas for shrinking models and inventive use of existing models than on simply increasing investment in compute resources.

Concerns over risks from generative artificial intelligence systems have increased significantly over the past year, driven in large part by the advent of increasingly capable large language models. But, how do AI developers attempt to control the outputs of these models? This primer outlines four commonly used techniques and explains why this objective is so challenging.