Memory safety issues remain endemic in cybersecurity and are often seen as a never-ending source of cyber vulnerabilities. Recently the topic has increased in prominence with the White House Office of the National Cyber Director (ONCD) releasing a request for comments on how to strengthen the open-source ecosystem. But what exactly is memory safety? This blog describes the historical antecedents in computing that helped create one aspect of today’s insecure cyber ecosystem. There will be no quick fixes, but there is encouraging progress towards addressing these long-standing security issues.

Universities can build more inclusive computer science programs by addressing the reasons that students may be deterred from pursuing the field. This blog post explores some of those reasons, features of CS education that cause them, and provides recommendations on how to design learning experiences that are safer and more exploratory for everyone.

The much-anticipated National Cyber Workforce and Education Strategy (NCWES) provides a comprehensive set of strategic objectives for training and producing more cyber talent by prioritizing and encouraging the development of more localized cyber ecosystems that serve the needs of a variety of communities rather than trying to prescribe a blanket policy. This is a much-needed and reinvigorated approach that understands the unavoidable inequities in both cyber education and workforce development, but provides strategies for mitigating them. In this blog post, we highlight key elements that could be easily overlooked.

Large language models (LLMs) could potentially be used by malicious actors to generate disinformation at scale. But how likely is this risk, and what types of economic incentives do propagandists actually face to turn to LLMs? New analysis uploaded to arXiv and summarized here suggests that it is all but certain that a well-run human-machine team that utilized existing LLMs (even open-source ones that are not cutting edge) would save a propagandist money on content generation relative to a human-only operation.

How can we measure the reliability of machine learning systems? And do these measures really help us predict real world performance? A recent study by the Stanford Intelligent Systems Laboratory, supported by CSET funding, provides new evidence that models may perform well on certain reliability metrics while still being unreliable in other ways. This blog post summarizes the study’s results, which suggest that policymakers and regulators should not think of “reliability” or “robustness” as a single, easy-to-measure property of an AI system. Instead, AI reliability requirements will need to consider which facets of reliability matter most for any given use case, and how those facets can be evaluated.

On July 21, the White House announced voluntary commitments from seven AI firms to ensure safe, secure, and transparent AI. CSET’s research provides important context to this discussion.

Recent discussions of AI have focused on safety, reliability, and other risks. Lost in this debate is the real need to secure AI against malicious actors. This blog post applies lessons from traditional cybersecurity to emerging AI-model risks.

Machine learning advances have powered the development of new and more powerful generative language models. These systems are increasingly able to write text at near human levels. In a new report, authors at CSET, OpenAI, and the Stanford Internet Observatory explore how language models could be misused for influence operations in the future, and they provide a framework for assessing potential mitigation strategies.