--- date: 2026-05-28 subject: "Illinois mandates AI safety audits | Trump AI EO talks continue under NDA | Dell wins $9.7B Pentagon deal" --- **Illinois lawmakers cleared SB 315 mandating third party safety audits** of frontier developers including OpenAI, Anthropic and Google DeepMind, with Gov. JB Pritzker saying he plans to sign. **White House negotiations over a frontier AI executive order** that President Trump publicly shelved May 21 have continued under nondisclosure agreements with the same three labs. **The administration is weighing direct financing for U.S. drone makers** including Neros, Performance Drone Works and Donald Trump Jr.-linked Unusual Machines, with some proposals contemplating government equity stakes. **A super PAC backed by OpenAI, Palantir and a16z executives has spent an estimated $2.4 million** against New York RAISE Act author Alex Bores in his NY-12 Democratic primary. # 1. AI Policy Today - **Illinois passes SB 315 requiring third party audits of OpenAI, Anthropic and Google frontier models** — The Illinois House of Representatives gave final approval to SB 315 on Wednesday, sending the bill to Gov. JB Pritzker, Wired reported. The measure requires frontier AI labs including OpenAI, Anthropic and Google DeepMind to have their safety practices audited by a third party. Pritzker said in a social media post Wednesday he plans to sign it, citing a need to hold Big Tech accountable. California and New York currently require AI safety reporting and incident disclosure; Illinois goes further by mandating independent verification that a lab is following its own safety commitments. OpenAI's Chief Global Affairs Officer Chris Lehane endorsed SB 315 in a statement to Wired. The Chamber of Progress, which counts Google and Andreessen Horowitz as partners, wrote to lawmakers Wednesday morning asking them to oppose. [Wired](https://www.wired.com/story/illinois-pass-major-ai-safety-law-pritzker/) - **Trump frontier AI executive order remains in talks under NDAs with OpenAI, Anthropic and Google** — The White House frontier AI executive order President Trump publicly called off May 21 has continued in negotiations under nondisclosure agreements with OpenAI, Anthropic and Google over recent weeks, POLITICO Pro reported. Last minute calls from Elon Musk, Mark Zuckerberg and former AI and crypto czar David Sacks persuaded Trump to halt the planned signing ceremony, the Washington Post reported. Musk denied the account on X, calling the report false and saying Trump spoke with him only after declining to sign, while Meta said Zuckerberg spoke with the president only after the order had been rescinded. Sacks warned the proposed voluntary system could function as a de facto mandatory regime requiring government green lights for model releases. The draft order would have created a voluntary system for companies to share frontier models with the government up to 90 days before public release, allowing agencies to test for dangerous capabilities. A federal official told the Post the order "is not dead" and is likely to be revisited. [POLITICO Pro](https://subscriber.politicopro.com/article/2026/05/it-isnt-canceled-inside-the-white-house-divisions-on-ai-00938557) [Washington Post](https://www.washingtonpost.com/politics/2026/05/22/last-minute-lobbying-by-tech-industry-officials-led-trump-cancel-ai-order/)[Yahoo](https://www.yahoo.com/news/politics/articles/killed-trumps-ai-order-musk-182846358.html) - **Pentagon awards Dell Federal Systems $9.7 billion contract to consolidate Microsoft software licenses across DOD** — The Pentagon awarded Dell Federal Systems a $9.7 billion Core Enterprise Technology Agreement to consolidate Microsoft software, services and licenses across the Department of Defense, the Intelligence Community and Coast Guard, CNBC reported. DOD chief information officer Kirsten Davies told reporters the five year blanket purchase agreement supports the department's pivot to AI and data analytics, with consolidation expected to save $422 million annually, per Breaking Defense. The award follows months of Dell engagement with the White House, including President Trump previously urging federal purchases of Dell computers. The agreement is not new funding but consolidates existing IT budgets managed by the Navy. [CNBC](https://www.cnbc.com/2026/05/27/dell-dod-pentagon-software-deal-digital-infrastructure-trump.html) [Breaking Defense](https://breakingdefense.com/2026/05/pentagon-awards-dell-9-7-billion-contract-to-consolidate-software-licenses/) - **Trump administration in talks to fund U.S. drone makers, including Donald Trump Jr.-linked Unusual Machines** — The Trump administration is in talks to provide funding to U.S. drone companies including Sequoia Capital-backed Neros and Unusual Machines, the Wall Street Journal reported. Unusual Machines is a drone components maker that counts Donald Trump Jr. as an adviser, while Neros is a startup specializing in autonomous drones. Performance Drone Works, which won a contract to supply the U.S. Army with reconnaissance drones, is also under consideration. Months-long Pentagon discussions have included the Office of Strategic Capital, a Biden era lending unit focused on companies critical to national security supply chains. Some proposals envision a mix of debt and equity that could give the government ownership stakes. [WSJ](https://www.wsj.com/politics/national-security/trump-us-drone-company-funding-cadef1f7?mod=pls_whats_news_us_business_f) [Investing.com](https://www.investing.com/news/stock-market-news/trump-administration-in-talks-to-fund-us-drone-companies-wsj-reports-4713572) - **Sen. Warren proposes overhauling tax code to tax AI companies and data center energy use** — Sen. Elizabeth Warren (D-Mass.) called for overhauling the tax code to ensure Americans share in the economic gains of AI, in a Time op-ed published Wednesday. Warren proposes a wealth tax on billionaires including Jeff Bezos and Sam Altman and a "reasonable" excise tax on data center energy consumption scaled to operations, Axios reported. The Massachusetts Democrat said the current code penalizes hiring through payroll taxes while subsidizing investments in technology. She called for raising taxes on corporations and capital gains and closing corporate loopholes. Warren cited a 267% rise in electricity costs in some communities over the past five years driven in part by data center power demands. [The Hill](https://thehill.com/homenews/senate/5898081-warren-proposes-ai-tax/) [Axios](https://www.axios.com/2026/05/27/elizabeth-warren-tax-ai-companies-benefit-americans) - **AFT President Weingarten proposes screen ban through grade 2 and student-facing AI ban in elementary schools** — American Federation of Teachers President Randi Weingarten on Wednesday called for a ban on screens for students in preschool through second grade and a prohibition on student-facing AI tools in elementary schools, the New York Times reported. AFT, which represents 1.8 million educators, framed the proposals as part of a 10 point plan delivered at the National Press Club in Washington. The plan would also ban social companion chatbots until age 16 and create an independent research consortium to study AI and screen effects on student learning, AOL noted. Weingarten told NBC News she views many of the big technology companies as "playing a really negative role" in pushing more tech into schools. K-12 Dive added that AFT entered a $23 million partnership with Microsoft, OpenAI and Anthropic last year for the National Academy for AI Instruction. [NYT](https://www.nytimes.com/2026/05/27/technology/ai-screens-schools-weingarten.html) [K-12 Dive](https://www.k12dive.com/news/aft-president-urges-bans-on-screens-student-facing-ai-for-youngest-learner/821263/) [AOL](https://www.aol.com/articles/teachers-union-president-calls-limits-130000647.html) - **ICE expands iris recognition contract with BI2 Technologies above $25 million, more than 5x prior year** — Immigration and Customs Enforcement is spending more than five times last year on a single vendor's identity verification technology, FedScoop reported, citing procurement documents published this month. ICE's prior contract with Massachusetts based BI2 Technologies ran $4.6 million from September 2025 to September 2026; the new award, running June 1, 2026 to May 31, 2027, surpasses $25 million. BI2 will supply 1,570 additional wireless iris-scanning handhelds linked to its Inmate Identification and Recognition System, which holds more than 5 million booking records from 47 states. The DHS justification cited multimodal biometrics combining iris, fingerprint and facial recognition, and noted BI2 has no public plans for FedRAMP authorization despite the sensitive data involved. [FedScoop](https://fedscoop.com/dhs-ice-agents-biometrics-contract-iris-recognition/) - **OpenAI- and a16z-aligned super PAC spending hits $2.4 million against NY-12 candidate and RAISE Act author Alex Bores** — Leading the Future, a super PAC funded by OpenAI, Palantir and a16z executives, has spent an estimated $2.4 million attacking New York state assemblyman Alex Bores in the NY-12 Democratic congressional primary since December 2025, The Verge reported. Bores authored the New York RAISE Act, signed into law in December 2025, which placed restrictions on the release of frontier AI models. A new Emerson College poll released last week shows him neck-and-neck with rival Micah Lasher, trailing by two points. Bores leads the broader eight person field in other recent polling. The Bores campaign placed its first paid New York ad on May 11, nearly seven months after entering the race; the primary is June 23. [The Verge](https://www.theverge.com/policy/937650/ai-alex-bores-openai-anthropic-ny12) # 2. China Watch - **Chinese researchers detail AI agent that autonomously designs bearings for defense machinery** — Researchers at Chongqing University's State Key Laboratory of Mechanical Transmission for Advanced Equipment built an AI bearing design agent that generates rolling bearings used in advanced machinery, the South China Morning Post reported. The system, called ChatBearing, combines large language models with engineering calculation tools and industrial databases. It was detailed in a paper in Acta Armamentarii, a Chinese defense engineering journal covering aerospace, missile systems, armor and guidance for the country's weapons industry. [South China Morning Post](https://www.scmp.com/news/china/science/article/3354989/ai-massively-increasing-chinas-new-weapon-development-speed-scientists?utm_source=rss_feed) - **Moffett AI raises nearly 1 billion yuan ($140 million) Series C for sparse computing AI chips** — Chinese AI chip startup Moffett AI raised nearly 1 billion yuan ($140 million) in Series C funding to commercialize its SparsePrime computing card, Pandaily reported. Moffett develops sparse computing technology, which strips algorithmic redundancy at run time to lift throughput on neural network workloads. The financing lands amid a wave of late stage rounds for domestic AI accelerator startups as Beijing pushes substitution for Nvidia parts blocked by U.S. export controls. [Pandaily](https://pandaily.com/moffett-ai-raises-nearly-1-billion-yuan-in-series-c-for-sparse-computing-ai-chips) # 3. Capability & Research Watch - **Cisco AI Threat Research finds multi-turn attacks succeed against all 15 closed frontier models tested** — Cisco AI Threat Research published a cross-vendor study finding all 15 closed frontier AI models tested were vulnerable to multi-turn adversarial prompting, SiliconANGLE reported. Success rates ranged 7.9% to 88.3% across 15 proprietary models from OpenAI, Anthropic, Google, Amazon and xAI. xAI's Grok 4.1 Fast in non-reasoning configuration showed the widest jump, moving from 34.2% single-turn to 88.3% multi-turn. Google's Gemini 3 Pro rose from 18.1% to 73.4%; Anthropic's Claude Opus 4.5 moved from 2.19% to 11.2%, the narrowest gap. The evaluation drew on 30,090 single-turn prompts and 6,986 multi-turn attacks across 1,456 conversations under Cisco's Integrated AI Security and Safety Framework. Cisco called on model providers to document the safety effects of configuration flags such as reasoning modes, system prompt adherence and guardrail tiers alongside capability benchmarks. [SiliconANGLE](https://siliconangle.com/2026/05/27/cisco-report-finds-no-closed-frontier-ai-model-safe-multi-turn-attacks/) - **RUSI warns North Korea and Iran moving from AI assisted to AI enabled sanctions evasion** — A new paper from the Royal United Services Institute (RUSI), "Algorithms of Evasion: The Rise of AI-Enabled Proliferation Financing," warns that North Korea and Iran are developing and deploying AI models to aid sanctions evasion, CSO Online reported. The report says AI can now mass-produce high quality fraudulent identification documents and automate "the administrative minutia of managing extensive shell company networks." AI powered systems analyze blockchain patterns in real time to adjust cryptocurrency mixing strategies. Author Aaron Arnold of RUSI's Centre for Finance and Security said in an email that the paper was prompted by a yearlong uptick in North Korea's use of AI in phishing schemes to fund ballistic missile and nuclear weapons programs. The paper estimates governments and the private sector have three to five years to adapt identification and mitigation protocols. [RUSI](https://www.rusi.org/explore-our-research/publications/research-papers/algorithms-evasion-rise-ai-enabled-proliferation-financing)[CSO Online](https://www.csoonline.com/article/4177936/another-it-governance-headache-ai-enabled-sanction-evasion-3.html) # 4. Industry & Market Watch - **NewsGuild files unfair labor practice charge alleging NYT used AI to monitor unionized tech workers** — NewsGuild of New York filed two grievances and an unfair labor practice charge accusing The New York Times of using AI to monitor and surveil performance of unionized tech workers in violation of their collective bargaining agreement, Editor & Publisher reported. The Tech Guild alleges the company deployed internal AI tools (including engineering productivity tool DX and knowledge search tool Glean) to track and evaluate employees without providing required information to the union, per The Verge. A Times spokesperson said the company disagrees with the characterizations in the grievances and will respond as part of its normal contractual process. Tech Guild AI committee chair Ben Harnett said using AI to surveil work creates "a skewed, inaccurate picture of our members' work." [Editor & Publisher](http://www.editorandpublisher.com/stories/newsguild-of-ny-tech-guild-take-legal-action-against-the-new-york-times,261833) [The Verge](https://www.theverge.com/ai-artificial-intelligence/937689/new-york-times-tech-guild-ai-monitoring-performance-union-contract) - **OpenAI Foundation commits $250 million for AI workforce disruption programs and research** — The OpenAI Foundation, the nonprofit controlling OpenAI, committed an initial $250 million for grants, partnerships and direct work to help workers and economies adapt to AI disruption, per IBTimes. The funds will back research into AI's labor market impact, support workers facing near term displacement and test new approaches to distributing AI's economic gains, per Reuters. The commitment is the foundation's first major spend on workforce disruption issues and part of a previously announced $1 billion 12 month plan. The Foundation received a 26% stake in the OpenAI for-profit entity last year as part of the restructuring; its holding was valued at $130 billion at the time. [IBTimes](https://www.ibtimes.com/openai-foundation-launches-250-million-initiative-address-ai-workforce-disruption-3803387) [Investing.com](https://www.investing.com/news/stock-market-news/openai-foundation-commits-250-million-to-help-workers-economies-navigate-ai-disruption-4712685) - **Robinhood opens platform to AI agents for stock trading and credit card payments** — Robinhood launched Agentic Trading and an Agentic Credit Card allowing AI agents to trade stocks and make credit card purchases on customers' behalf, the company said in a Wednesday news release covered by Forbes. Users create a separate account for their AI agents and connect them to a dedicated wallet. Agents can analyze portfolios for concentration risk and sector exposure, execute trades and review analyst notes, with users setting monthly limits and approval requirements, per TechCrunch. The agentic credit card is initially available only to Robinhood Gold Card holders. The platform uses Model Context Protocol servers for connectivity, with fraud detection flagging suspicious agent trades for human review. [Forbes](https://www.forbes.com/sites/ronschmelzer/2026/05/27/when-ai-agents-get-your-wallet/) [TechCrunch](https://techcrunch.com/2026/05/27/robinhood-now-lets-your-ai-agents-trade-stocks/) - **YouTube moves AI generated content labels to more prominent positions and rolls out automatic identification** — YouTube is relocating AI generated content labels to more prominent positions on Shorts and long form videos, The Verge reported. The platform is also rolling out automatic AI content identification, applying labels without requiring creator self-disclosure. The changes coincide with Google's expanded AI verification efforts unveiled at this year's I/O developer conference. [The Verge](https://www.theverge.com/streaming/937915/youtube-ai-labels-shorts-automatic-identification-updates) # 5. Global & Geopolitics - **EU draft "tech sovereignty" strategy proposes restricting U.S. cloud services for sensitive public data** — The European Commission is preparing a draft tech sovereignty strategy that says the bloc must "reclaim its place in the global race for geoeconomic power" by backing European alternatives in semiconductors, cloud computing and AI, the Financial Times reported. The package, slated for May 27 presentation, would consider rules restricting the use of cloud services from non-EU countries for sensitive public data in healthcare, finance and the judiciary, per CNBC sources. The plan includes incentives to accelerate European data center construction and favor homegrown cloud and AI technologies. EU leaders will weigh the package and additional supplier diversification rules at a Friday meeting in Brussels. [FT](https://www.ft.com/content/c05e152d-36e5-4446-a706-5f5fa5d98315) [CNBC](https://www.cnbc.com/2026/05/07/eu-commission-cloud-sensitive-data.html) - **EU executive arm to back tougher line on Chinese imports with supplier diversification rules at Friday meeting** — The EU executive arm is preparing to back a tougher line against Chinese imports at a Friday meeting in Brussels, South China Morning Post reported. Among potential measures: requiring at least three suppliers from two different countries in key sectors. The industry commissioner has spoken publicly about the diversification framework. [SCMP](https://www.scmp.com/plus/news/china/diplomacy/article/3355161/europe-plans-tougher-response-china-shock-20?utm_source=rss_feed) - **Netherlands blocks Kyndryl's $113 million acquisition of cloud provider Solvinity on national security grounds** — The Dutch government blocked New York-based Kyndryl's €100 million ($113 million) acquisition of Dutch cloud provider Solvinity on Tuesday, Investing.com reported. It is the first U.S. acquisition the Dutch Investment Screening Bureau has rejected since its 2020 establishment. Dutch minister for the digital economy Willemijn Aerdts said in a published letter the government has imposed a "complete prohibition" on the deal, per TechCrunch. Solvinity hosts DigiD, the Dutch government identity platform used by residents to access public services. The U.S. Embassy in The Hague said it was "disappointed," The Hill reported. [The Hill](https://thehill.com/homenews/5898407-dutch-block-us-tech-acquisition/) [TechCrunch](https://techcrunch.com/2026/05/26/dutch-government-blocks-us-company-from-acquisition-citing-risk-to-public-interest/) [Investing.com](https://uk.investing.com/news/stock-market-news/dutch-government-blocks-kyndryls-113-million-solvinity-acquisition-93CH-4698434) - **Google's $15 billion Visakhapatnam data center hub draws $2.3 billion in Andhra Pradesh subsidies as locals face water shortages** — Andhra Pradesh is extending roughly $2.3 billion in subsidies over 20 years to Google's $15 billion data-center project in Visakhapatnam, including a 25% discount on water and land for 10 years, the Wall Street Journal reported. The 600-acre, three-site hub will be India's first focused on training large-scale AI models for a U.S. tech firm; at full tilt it would draw the annual electricity equivalent of six million Indians. The World Resources Institute classifies the city of two million as under "extremely high" water stress, with some areas getting less than an hour of tap water a day. Farmers pushed off mango and cashew plots are getting up to 4 million rupees ($42,000) per acre. [WSJ](https://www.wsj.com/tech/ai/big-subsidies-for-google-limited-water-for-locals-the-dilemma-of-ai-in-india-105a770e?mod=rss_Technology)