Center for Security and Emerging Technology (CSET)

?? New blog ?? How do we make AI #redteaming more useful for more people? Jessica J. draws from a CSET workshop on AI testing to lay out the challenges and how we can tackle them. ?? Learn more: https://lnkd.in/eGSn3sXp

My latest op-ed in Breaking Defense is out today. In it, I discuss the recent termination of the Advisory Committee on Excellence in Space (ACES) under the Office of Space Commerce. ACES was a crucial bridge between government, industry, and academia, ensuring that the U.S. #space sector remained competitive, innovative, and secure. By shutting it down, the administration is weakening the very public-private collaboration that has fueled American leadership in space. We’ve already seen how indispensable commercial space technology has been in global security, particularly in Ukraine. Companies like #SpaceX, #Maxar, and #PlanetLabs have demonstrated the power of commercial innovation in #national #defense. Yet, instead of strengthening these ties, the U.S. government is now making it even harder for industry to engage, limiting its own ability to maintain technological superiority. This move contradicts previous U.S. space #policy from the first Trump administration, which emphasized the importance of a robust and competitive #commercialspace sector. It also raises questions about whether this administration is shifting toward favoring a select few dominant players, a strategy that could stifle competition and innovation in the long run. The elimination of ACES is more than an administrative change—it’s a setback for #nationalsecurity, economic growth, and U.S. leadership in space. The administration should reconsider. The stakes are simply too high. Read my full take here: https://lnkd.in/eD9jSH2K

NEW: In this month’s newsletter, CSET's Alex Friedland wrapped up all the latest AI & policy news: ?? GPT-4.5 & Claude 3.7 Sonnet — OpenAI and Anthropic try to stay at the head of the pack ?? Recapping new misalignment research ?? SecDef Hegseth's new memo to speed up DOD software acquisition ?? AI Action Plan recs from CSET, OpenAI, Anthropic, Google, IFP – Institute for Progress, Center for a New American Security (CNAS), and others And much more! ?? Check it out and subscribe: https://lnkd.in/eJDYsP6r

New short paper from Kendrea Beers and me - 2 case studies of OpenMined's great work giving researchers/auditors/etc access to test privately held AI models. In pilot 1 DailyMotion, a French video sharing site, connected OpenMined's privacy-preserving infrastructure to their stack. This let an external researcher analyze what kind of content they were upranking *without* needing access to user data or the algorithm ?? Pilot 2 was even cooler: Anthropic & UK AISI did an exercise in "mutual secrecy," using OpenMined's tech to run a biosecurity evaluation where the AI model was kept private from the UK govt and the biological dataset was private from Anthropic. Just a test exercise with toy model and dataset so far, but the idea is that next this could be used to allow them to test a sensitive model on a sensitive dataset - without giving access to either. Still early days for this technology but it's cool stuff, and we're proud that Center for Security and Emerging Technology (CSET)'s Foundational Research Grants program helped to fund OpenMined's work on this. Preprint of the paper (which Kendrea presented at the Conference on Frontier AI Safety Frameworks): https://lnkd.in/e8XGMpUe Call for proposals on a related topic from Open Philanthropy: https://lnkd.in/ehcMJsqe More technical detail from OpenMined on the secure enclaves on H100s they used to make the Anthropic/UKAISI pilot happen: https://lnkd.in/eCmS6dpc

Happy to share that I'll be speaking at What’s Next for AI Red-Teaming, make sure to attend it on March 25.

? New CSET Report ? As new advanced AI systems roll out, there is widespread disagreement about malicious use risks. Are bad actors likely to misuse these tools for harm? Our new report by Josh Goldstein and Girish Sastry presents a simple framework to guide the questions researchers ask—and the tools they use—to evaluate the likelihood of malicious use. Read the full report here: https://lnkd.in/eqP2hDsX

Earlier today, the Center for Security and Emerging Technology (CSET) published our response to #OSTP's request for input on an #AI Action Plan. One recommendation that I personally contributed was that the U.S. government should develop and adopt standards to mitigate risks from AI. Specifically, AISI should develop standards on topics such as: 1?? Model training 2?? Pre-release internal & external security testing 3?? Cybersecurity practices 4?? If-then commitments 5?? AI risk assessments 6?? Processes for testing and re-testing systems as they change over time. These standards, if implemented, would go a long way towards mitigating the potential risks of AI and increasing public trust and confidence in using it, allowing us to realize its benefits sooner than we could otherwise. Check out the full response to the RFI here:

?? This is what the brightest minds in AI governance think the priorities for the US AI Action Plan should be. I might be biased, but my colleagues at CSET never fail to do great work, and this is another great example of the value of independent research.

Our whole team at the Center for Security and Emerging Technology (CSET) prepared a response to #OSTP's request for input on an #AI Action Plan. Our recommendations can be summarized into three major themes: 1?? To Secure and Advance U.S. AI Leadership 2?? To Navigate U.S.-China Technology Competition 3?? To Realize AI’s Benefits While Managing Risks Additionally, I add that maintaining U.S. competitiveness in AI requires developing and sustaining the necessary #workforce. These are absolutely mutually exclusive. Two things I'd love to see this administration prioritize are (a) scholarship-for-service programs that are direct talent pipelines into government service and (b) developing AI literacy for the American citizenry. To that end, I recommend: ? ?Supporting the creation of an AI scholarship-for-service program. In 2024, the NSF released a report detailing the feasibility of and need for an AI scholarship-for-service program following the CHIPS and Science Act. ? Work with Congress to support AI literacy efforts for the American people. In 2024, Senators Kelly and Rounds introduced a bill aimed at bolstering consumer awareness and confidence in the use of AI products and services. A companion bill was later introduced by Representative Blunt Rochester. (I am proud to have offered comments of support for both.) Read our response here ?? https://lnkd.in/eAEi7m4W

CSET’s response to the Trump administration’s Request for Information on the Development of an AI Action Plan was truly a team effort. The response focuses on steps the United States can take to advance AI leadership, remain competitive in AI on the global stage, and maximize AI’s benefits while minimizing its risks. We recommend that the administration support key enablers of U.S. technological prowess, including access to AI talent, foundational AI standards and evaluations, a rich information environment, protections from AI harms, and dynamic and open markets and research ecosystems. We also hope the RFI acts as a bridge to further exploring CSET’s rich areas of research. Check out the entire response here!