IDF uses AI to accelerate targeting according to report, the White House releases guidance on federal AI use, and a roundup of the latest computing and funding news

Sign up
Worth Knowing

IDF Accelerates Targeting with AI, According to Report: A report published last week by the Israeli outlets +972 Magazine and Local Call alleges that the Israel Defense Forces (IDF) have relied heavily on an AI-powered system — with minimal human oversight — to identify targets for airstrikes in Gaza. The report, which cites six unnamed Israeli intelligence officers and has not been independently verified (though the accounts were also shared with the Guardian), alleges that the system — dubbed “Lavender” — processed massive amounts of surveillance data to mark as many as 37,000 Palestinians as likely Hamas or Palestinian Islamic Jihad militants. According to the report, the IDF has come to rely heavily on Lavender’s lists since the early weeks of the current war in Gaza, with little practical human oversight. One source said they spent approximately 20 seconds evaluating each Lavender-selected target — only enough time to confirm the target was male. Another automated system identified by the report — referred to as “Where’s Daddy?” — has allegedly been used to track targets back to their private homes, which were then marked for bombing, presumably elevating the risk of civilian casualties significantly. In a lengthy response, the IDF disputed the report, writing that “the IDF does not use an artificial intelligence system that identifies terrorist operatives or tries to predict whether a person is a terrorist. Information systems are merely tools for analysts in the target identification process. … The ‘system’ … is not a system, but simply a database whose purpose is to cross-reference intelligence sources.”

Expert Take: “Many discussions of AI in warfare focus on the risks of lethal autonomous weapons systems (LAWS) that autonomously select their own targets. There’s often an assumption that keeping a human “in the loop” who has to approve targeting decisions solves the problem of LAWS. This report from Israel, if accurate, shows the flaw of not considering the human-machine team as a whole. What matters is not just whether a human is involved, but how they interact with the AI system in practice—which will depend on questions of design, operator training, and testing in realistic conditions. If those questions are not thought through carefully, the results can be disastrous regardless of whether the human or machine has final say.” — Helen Toner, Director of Strategy and Foundational Research Grants
140B Dollars and 208B Transistors — A Compute and Investment Roundup: The AI development landscape is increasingly awash in both computing power and the billions of dollars needed to pay for it. Notable examples from the last month include:
Government Updates

White House Debuts Guidelines for Federal Government AI Use: On March 28, Vice President Harris announced new policies governing the federal government’s use of AI. The guidelines, issued by the Office of Management and Budget (OMB), were required by President Biden’s October 2023 executive order on AI and set requirements for how federal agencies should deploy the technology, including:
  • By December 1, 2024, agencies are required to institute minimum practices when deploying “safety-impacting” and “rights-impacting” AI (both terms are defined in the memo). These practices include AI impact assessments, real-world testing, independent evaluation, ongoing monitoring, adequate training, and ample public notification and documentation. Agencies that cannot meet these minimum standards are required to stop using the system in question.
  • Agencies must conduct annual inventories of each AI use case and publish a public version of the inventory online. Beginning with this year’s inventories, agencies must identify safety-impacting and rights-impacting AI use cases, outline the risks posed, and explain how the agency is managing the risks. Not all agencies’ AI use cases fall under this requirement — the DOD is excepted, for example — but agencies still must report some aggregate metrics about their non-covered AI use cases.
  • Within 60 days of the guidance’s release, agencies must designate a Chief AI Officer to coordinate their agency’s use of AI, as initially mandated by the October 2023 executive order. Several agencies — including the Justice Department and the Department of Education — have already named CAIOs.
More AI news is still to come from OMB — concurrent with the release of the guidance, OMB issued a request for information on “Responsible Procurement of Artificial Intelligence in Government.” The deadline for submissions is April 29.

A Leadership Change at the Pentagon’s Lead AI Office: On Monday, Radha Plumb took over as the Department of Defense’s Chief Digital and Artificial Intelligence Officer (CDAO) from Craig Martell, who departed after leading the Pentagon’s AI office since 2022. First announced in 2021, the office of the CDAO combined several precursor organizations — including the Joint Artificial Intelligence Center — under one roof as part of the DOD’s effort to embrace AI. Martell, who came to DOD after a long career in the private sector, had a mixed tenure as the inaugural CDAO. There were some successes; in February, for example, Deputy Secretary of Defense Kathleen Hicks announced that the CDAO had helped deliver on one of its primary goals: to stand up a workable “minimum viable capability” of Combined Joint All-Domain Command and Control (CJADC2), the Pentagon’s ambitious plan to coordinate and connect its services’ sensors. But the office also suffered morale problems during Martell’s tenure — according to a 2023 DefenseScoop report, CDAO employees gave the office’s leadership negative ratings on an internal survey, the only negative rating in any category across the entire DOD. Unlike Martell, Plumb comes to the CDAO position from elsewhere in the DOD, having previously served as the Deputy Undersecretary for Defense Acquisition and Sustainment.

Commerce Announces Billions in Chips Act Funding: The Commerce Department has opened the spigots on CHIPS Act funding in recent weeks, announcing several preliminary funding agreements with major chipmakers, with more reported to be on the way. On March 20, the Commerce Department announced $8.5 billion in grants and as much as $11 billion in loans to Intel, which has plans to spend more than $100 billion over the next five years on projects in Arizona, New Mexico, Ohio, and Oregon. Then earlier this week, Commerce announced more than $11 billion — $6.6 billion in grants and $5 billion in loans — for TSMC. As we covered last month, concerns had begun to crop up around the company’s Arizona projects, but the funding announcement seems to have done quite a bit to heal any wounds; to coincide with the CHIPS-funding announcement, the Taiwan-based chipmaker announced plans for a third Arizona fab, bringing the price tag of its U.S. projects to more than $65 billion. And more major funding announcements are expected soon: Reuters reported on Monday that the Biden administration will award more than $6 billion to Samsung for its Texas-based projects — a previously announced $17 billion facility near Austin and $27 billion in soon-to-be-announced additional spending — and Bloomberg reported additional incoming grants for the Boise-based DRAM giant Micron in the coming weeks.

In Translation
CSET’s translations of significant foreign language documents on AI

PRC Generative AI Standard: Basic Safety Requirements for Generative Artificial Intelligence Services. This Chinese standard for generative AI establishes very specific oversight processes that Chinese AI companies must adopt in regard to their model training data, model-generated content, and more. The standard names more than 30 specific safety risks, some of which—algorithmic bias, disclosure of personally identifiable information, copyright infringement—are widely recognized internationally. Others, such as guidelines on how to answer questions about China’s political system and Chinese history, are specific to the tightly censored Chinese internet. One notable addition to this document, relative to a preliminary draft released in October 2023, is a clause requiring a supply chain security assessment of Chinese generative AI models’ underlying hardware and software.

PRC S&T Intelligence Center Document: International Science and Technology Information Center: Introduction to the Center. This document is a translation of the “About Us” page of the website of the International Science and Technology Information Center (ITIC), a government-run open-source S&T intelligence provider in Shenzhen, a tech hub in southern China. ITIC claims to use AI models to analyze all of the major global academic and corporate S&T literature databases so as to identify emerging trends in research and technology. The center makes its tools—notably its “Sci-Brain” or “S&T Supermind” platform—and underlying data available to the entire population of Shenzhen, free of charge.

PRC State-Run Enterprise Reform Document: Notice on the Initiative to Launch Value Creation Benchmarking Against World-Class Enterprises. This document, issued in 2022 by the Chinese government ministry in charge of state-run enterprises (SOEs) and popularly known simply as “Document 79,” announces a campaign to improve SOEs’ “value creation.” The Wall Street Journal reported that this directive requires SOEs in strategic economic sectors to replace all of their foreign IT software by 2027, a drive known informally in China as “Delete A,” for “delete America.” The publicly available version of Document 79 translated by CSET does not mention anything about removing foreign software, although it does emphasize the importance of SOEs improving their capacity for in-house innovation.

If you have a foreign-language document related to security and emerging technologies that you’d like translated into English, CSET may be able to help! Click here for details.

Job Openings

We’re hiring! Please apply or share the role below with candidates in your network:
  • Program Specialist: CSET is currently accepting applications for a Program Specialist position to oversee its intern and student portfolios. The Program Specialist will play a key role in enhancing and managing CSET’s Inaugural Internship Program. Reporting to CSET’s Director of People and Operations, the Program Specialist will oversee CSET’s student and internship initiatives, as well as provide general administrative, organizational, and project management support to ensure CSET’s operations run smoothly. Apply by Monday, May 6.
Please bookmark our careers page to stay up to date on all active job postings. You can also subscribe to receive job announcements by updating your subscription preferences in the footer of this email.

What’s New at CSET

REPORTS

PUBLICATIONS AND PODCAST APPEARANCES

EMERGING TECHNOLOGY OBSERVATORY

EVENT RECAPS

  • On April 10, CSET’s Josh A. Goldstein, Meta’s Dr. Sarah Shirazyan, the Brennan Center for Justice’s Lawrence Norden, and Dr. Jon Roozenbeek of the University of Cambridge discussed how AI could impact elections in 2024 and how policymakers, the media, AI providers, and social media companies can respond.

IN THE NEWS

What We’re Reading

Paper: Governing AI Agents, Noam Kolt (April 2024)

Report: Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Services Sector, U.S. Department of the Treasury (March 2024)

Upcoming Events


What else is going on? Suggest stories, documents to translate & upcoming events here.