AI Policy · Daily

Trump is expected to sign an AI cybersecurity executive order Thursday, establishing a voluntary 90-day pre-release framework for developers to share models with the government before launch. The Pentagon created a task force to integrate OpenAI and Google frontier models into U.S. Cyber Command and National Security Agency operations. Trump signaled that he would speak directly with Taiwan's President Lai Ching-te, which would be the first such call by a sitting U.S. president since 1979. War Department CTO Emil Michael told senators AI use jumped 1,775% over the past year, adding roughly 1.42 million users across the department.

I.AI Policy Today

Trump set to sign AI cybersecurity executive order Thursday with voluntary 90-day pre-release framework

President Trump is expected to sign an executive order on AI and cybersecurity as soon as today. The White House is working to bring AI company chief executives to a signing ceremony, two sources familiar with the matter told Reuters. The order would create a voluntary framework asking developers to provide their models to the government 90 days before public release, and to give critical infrastructure providers such as banks pre-public access. Pressure has come in part from MAGA activists including former Trump adviser Steve Bannon and political organizer Amy Kremer, who have urged mandatory government security tests. Venture capitalist Marc Andreessen and former Trump AI adviser David Sacks have resisted binding requirements. The directive advances the FDA-style pre-release framework that WSJ documented in administration deliberations, as reported in AIPD's May 14 edition.

Read at Bloomberg ↗ • Read at Investing.com ↗ • Read at Reuters ↗

Trump indicates direct call with Taiwan's Lai, the first such contact since 1979 diplomatic shift

President Trump again indicated he would speak directly with Taiwanese President Lai Ching-te, a step Bloomberg said would risk disrupting newly stabilized U.S.-China relations. Lai responded that if given the chance to speak with Trump he would tell him China was undermining peace and that no one had the right to "annex" the island, per Reuters in coverage carried by Investing.com. A presidential level call has not taken place since Washington shifted diplomatic recognition to Beijing from Taipei in 1979. Trump said last week, following his summit with China's Xi Jinping, that he was undecided on further arms sales to Taiwan and called them a "good negotiating chip."

Read at Bloomberg ↗ • Read at Investing.com ↗

Pentagon stands up task force to deploy OpenAI and Google frontier models across Cyber Command and NSA missions

The Pentagon launched a new task force charged with determining how AI models from companies including OpenAI and Google can be safely deployed across U.S. Cyber Command and National Security Agency missions, Politico reported. The initiative aims to operationalize commercial frontier AI inside the U.S. government's most sensitive cyber and signals intelligence networks. Separately, Anthropic's Mythos has been in use at the NSA for vulnerability scanning despite the Pentagon's pending supply chain risk designation against the company, Axios previously reported.

Read at Politico ↗ • Read at Axios ↗

FTC sends warning letters to a dozen websites on TAKE IT DOWN Act compliance

The Federal Trade Commission sent warning letters to a dozen websites advising them of their obligation to comply with the TAKE IT DOWN Act, the agency said in a press release. The law requires covered platforms to provide a mechanism for users to request removal of nonconsensual intimate imagery, including AI generated deepfakes, and to remove the imagery within 48 hours of a valid request. The new letters follow the FTC's May 19 enforcement opening targeting 15 named platforms, including Amazon, Alphabet, Apple, Meta, Microsoft, Reddit, Snapchat and TikTok, as reported in AIPD's May 20 edition.

Read at FTC ↗

War Department CTO says departmental AI use rose 1,775% in the past year, adding 1.42 million users

Use of artificial intelligence across the War Department is up by roughly 1.42 million users, or about 1,775%, over the past calendar year, Undersecretary of War for Research and Engineering Emil Michael said, per a War Department news release. Michael also serves as the department's chief technology officer. He testified May 19 before the Senate Armed Services Committee's emerging threats and capabilities subcommittee on the fiscal 2027 budget request, alongside science and technology leads from the Army, Navy and Air Force, per a Defense Visual Information Distribution Service post.

Read at DOD ↗ • Read at DVIDS ↗

Airbnb CEO Chesky defends Qwen use, says U.S. House probe misunderstands open source AI

Airbnb Chief Executive Brian Chesky defended the company's use of Chinese AI models in a Bloomberg TV interview, saying U.S. lawmakers worrying Chinese firms can access Americans' user data are "misunderstanding" the technology, per The Business Times. "We are not providing data to any Chinese companies," Chesky said, adding that Airbnb primarily uses a variety of open source models, including U.S. open source models. The remarks were Chesky's first public response to an April 29 House inquiry from the Homeland Security and Select China committees into Airbnb's use of Alibaba Group's Qwen large language model for its customer service chatbot, Bloomberg Law reported. The committees also sent a similar information request to Anysphere, the company behind the Cursor AI coding tool. Chesky said Airbnb is cooperating with the congressional committees, per Forbes.

Read at Forbes ↗ • Read at The Business Times ↗ • Read at Bloomberg Law ↗

II.China Watch

Alibaba unveils T-Head Zhenwu M890 AI chip and Qwen3.7-Max flagship at cloud summit

Alibaba on May 20 unveiled the Zhenwu M890, a new self-developed AI training and inference chip from its T-Head subsidiary, alongside the flagship Qwen3.7-Max large language model at the 2026 Alibaba Cloud Summit, Caixin reported. The chip delivers up to three times the performance of the prior generation, with 144 GB of memory and 800 GB per second of interchip bandwidth. Alibaba Cloud said its average daily token revenue has grown 15-fold this year and its model as a service offering has turned profitable. The company is capturing the bulk of incremental AI cloud, LLM and inference demand in finance, automotive, government and telecom sectors. The "national chip, national model" launch follows Beijing's recent ban on a key Nvidia accelerator and pending U.S. AI chip export controls.

Read at Caixin ↗

360 vulnerability platform joins MIIT's first filed list under China's national vulnerability management system

China's Ministry of Industry and Information Technology this week published its first list of 36 network product security vulnerability collection platforms cleared under the country's 2023 administrative measures, PingWest reported. 360 Digital Security Group's 360 Vulnerability Cloud Platform was among those granted filing status. This is the first time MIIT has publicly named filed platforms since the rules took effect. Filed platforms are formally incorporated into China's national vulnerability disclosure and management regime, which channels reports to MIIT and restricts overseas disclosure of certain flaws.

Read at PingWest ↗

DeepSeek forms Harness team to build standalone coding agent product

DeepSeek has set up a new Beijing-based Harness team focused on coding agents and posted listings for a product manager and an engineer, Pandaily reported. The job descriptions ask for heavy users of Claude Code, Cursor, Codex or GitHub Copilot. DeepSeek's V3 and R1 models already power a class of low-cost coding agent stacks popular with developers worldwide.

Read at Pandaily ↗

Tencent launches Mavis system level AI assistant with multi-agent orchestration and local privacy mode

Tencent on May 20 launched Mavis, a system level AI assistant available for Windows, Mac and Android with no invitation code required, TechNode reported. Mavis ships with a primary orchestration agent and five specialized agents for files, computer control, applications, browsing and search, executing tasks across the local file system and across devices. The product supports switching between edge and cloud compute and includes system level access to documents, computing resources and applications.

Read at TechNode ↗

III.Federal Policy Tracker

U.S. Army's JIATF 401 expands counter-drone marketplace to Australia, Poland and South Korea

Army Secretary Dan Driscoll signed agreements with counterparts from Australia, Poland and South Korea enabling each country to procure counter-small unmanned aerial system technologies through the U.S. Army-led Joint Interagency Task Force 401 drone defense marketplace, the War Department said. The marketplace offers low-collateral defeat effectors, radars, sensors and electronic warfare platforms, per Executive Gov. The South Korea letter of intent was signed by Patrick Mason, U.S. Army deputy assistant secretary for defense exports and cooperation, and Jeon Joon-boem, director general of the South Korean Defense Ministry's artificial intelligence planning bureau, Stars and Stripes reported.

Read at DOD ↗ • Read at Executive Gov ↗ • Read at Stars and Stripes ↗

IV.Industry & Market Watch

AMD pledges over $10 billion in Taiwan AI investments to expand packaging and partnerships

Advanced Micro Devices pledged to invest more than $10 billion in Taiwan to expand partnerships and add packaging capacity, Bloomberg reported. The investments will focus on partnerships to advance chip packaging and manufacturing required for next-generation AI infrastructure, per CNBC. The U.S. chipmaker plans to tap Taiwan's high-performance computing expertise, per WSJ.

Read at Bloomberg ↗ • Read at CNBC ↗ • Read at WSJ ↗

Box Elder County, Utah, approved O'Leary-backed Stratos data center projected to consume 9GW

Box Elder County commissioners signed off on the Stratos Project, a 40,000-acre data center stretching across the county's Hansel Valley, on May 4, The Verge reported. Backed by Shark Tank investor and venture capitalist Kevin O'Leary, the project is projected to be more than twice the size of Manhattan and to consume 9 GW of power. The approval has drawn pushback from experts and local residents citing environmental damage and strain on water supplies.

Read at The Verge ↗

OpenAI plans confidential IPO filing in coming weeks, targeting September listing

OpenAI is preparing to confidentially file for a U.S. initial public offering in the coming weeks and is aiming to list as soon as September, the Financial Times reported. The Sam Altman-led company is working with Goldman Sachs and Morgan Stanley on a draft IPO prospectus, Reuters added in coverage carried by Investing.com. The filing comes two days after OpenAI fended off an existential court challenge from Elon Musk, with Reuters earlier reporting the company had been laying the groundwork for an offering that could value it at up to $1 trillion.

Read at Financial Times ↗ • Read at Investing.com ↗

Samsung averts 18-day strike with tentative wage deal as government threatened emergency powers

Samsung Electronics' labor union said it would hold off on a planned 18-day strike and put a tentative wage deal with management to a vote May 22-27, AP reported. The agreement came after government mediated negotiations over how much of the AI fueled profit boom should pass to chip workers as bonuses. South Korean officials had earlier threatened to invoke rarely used emergency powers to force a settlement at the world's largest memory chip maker, the Straits Times reported. The union represents more than 90,000 Samsung workers, and Samsung accounts for more than 20% of South Korean exports.

Read at Financial Times ↗ • Read at AP ↗ • Read at Straits Times ↗

V.Global & Geopolitics

Demos study finds AI chatbots gave voters misinformation 34% of the time before Scottish election; UK Electoral Commission seeks controls

An investigation by UK think tank Demos found AI services gave voters misinformation in response to 34% of test questions posed during the recent Scottish election, The Guardian reported. The chatbots produced fabricated scandals, nonexistent candidates and incorrect dates. The study tested ChatGPT, Replika, Grok and Gemini. The UK Electoral Commission has called for new legal controls over chatbot misinformation in response.

Read at The Guardian ↗

EU shortlists tungsten, rare earths and gallium for first joint critical minerals stockpile, citing China dependence

The European Union has shortlisted tungsten, rare earths and gallium for its first joint stockpile of critical minerals, aiming to reduce reliance on China for materials vital to defense, semiconductors and the energy transition, the South China Morning Post reported. The bloc is in talks with major ports including Rotterdam in the Netherlands to store the minerals. Magnesium is also expected to make the priority list, with germanium and graphite likely to be added, Reuters reported in coverage carried by Mining Weekly. Most minerals on the list appear on NATO's list of 12 elements deemed critical to the defense industry.

Read at South China Morning Post ↗ • Read at Mining Weekly ↗

Google DeepMind agrees to Acas talks with UK unions over staff concerns on U.S. and Israeli AI defense use

Google DeepMind agreed to enter formal talks with the Communications Workers Union and Unite at the UK's Advisory, Conciliation and Arbitration Service after London staff voted to pursue trade union representation, The Guardian reported in an exclusive. The Acas talks follow petitions signed by workers about the U.S. and Israeli governments' use of its AI for defense and intelligence. The unionization drive at DeepMind's London headquarters is the first formal labor push at the lab, which is led by Nobel laureate Demis Hassabis.

Read at The Guardian ↗