AI Policy · Daily

Five Eyes cybersecurity agencies told leaders to brace for frontier AI cyberattacks within months, in a joint statement from the U.S., UK, Canada, Australia and New Zealand. Manhattan Democrats vote today in NY-12, where state Assemblyman Alex Bores faces roughly $8 million in attack ads from an OpenAI linked super PAC and $9 million in support from an Anthropic affiliated one. OpenAI rolled out Daybreak, a cybersecurity suite built around a new GPT-5.5-Cyber model and a "Patch the Planet" program aimed at open source vulnerabilities. Oracle cuts 21,000 jobs in one year, citing AI adoption.

I.Top Stories

Five Eyes cybersecurity agencies warn AI cyberattacks could land within months

Cybersecurity agencies of the United States, the United Kingdom, Canada, Australia and New Zealand issued a joint statement late Monday warning that frontier AI models capable of escalating cyber operations are months from broad accessibility, per The Guardian. The agencies told leaders to "act now" and said AI accelerates the speed, scale and sophistication of cyberthreats even as it aids defense. The statement said advanced models could exceed current industry expectations and reshape attack and defense capabilities.

Read at The Guardian ↗ • Read at FT ↗ • Read at National Cyber Security Centre ↗

AI super PAC spending faces ballot test in Manhattan's NY-12 Democratic primary today

Manhattan Democrats vote today in the open NY-12 House primary to succeed retiring Rep. Jerry Nadler, the race that has drawn heavy AI industry super PAC spending this cycle, per Axios. State Assemblyman Alex Bores, author of New York's RAISE Act on frontier AI safety, faces nearly $8 million in opposition spending from the OpenAI connected Think Big PAC and roughly $9 million in support from the Anthropic affiliated Jobs and Democracy PAC. Half of AI super PAC ad money this cycle has gone to the single Manhattan race, The Guardian reported. Mayor Zohran Mamdani is separately backing former Comptroller Brad Lander against Rep. Dan Goldman in NY-10 and Darializa Avila Chevalier against Rep. Adriano Espaillat in NY-13.

Read at Axios ↗ • Read at The Guardian ↗

OpenAI launches Daybreak cybersecurity suite with GPT-5.5-Cyber and 'Patch the Planet'

OpenAI introduced its Daybreak suite Monday, packaging Codex Security and an upgraded GPT-5.5-Cyber model designed to help organizations find, validate and patch software vulnerabilities. Wired reported the launch was pitched as a competitor to Anthropic's Mythos cybersecurity push. The package includes "Patch the Planet," an initiative aimed at supporting open source software maintainers with AI vulnerability discovery and expert review.

Read at Wired ↗ • Read at OpenAI ↗

Oracle cuts 21,000 jobs in 12 months, citing AI adoption in workforce reductions

Oracle reduced its global workforce by 21,000 employees in the past 12 months, to 141,000 as of May 31, with the company saying in an annual filing that "the adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce," per Bloomberg. The reductions led to about $1.8 billion in restructuring costs. Oracle ended the period with about 49,000 U.S. workers and roughly 92,000 internationally, just below its headcount before the 2022 Cerner acquisition. The cuts come as Oracle invests heavily in building AI data centers for customers including OpenAI.

Read at Bloomberg ↗ • Read at CNBC ↗

II.China Watch

Tencent begins testing a native AI assistant inside WeChat

Tencent has begun a closed beta of Xiaowei, a native AI assistant embedded into WeChat, the messaging platform serving 1.4 billion users, per Pandaily. The assistant runs on Tencent's in-house WeLM large language model, with DeepSeek supplementing some responses. It can set reminders, send WeChat messages, summarize forwarded documents, operate Meituan and JD.com mini-programs on a user's behalf, and generate custom mini-programs from natural language descriptions. Tencent is targeting a wider public rollout in the third quarter. The beta falls under the Cyberspace Administration of China's generative AI registration regime, covering agent payments, mini-program code generation and chat data handling.

Read at Pandaily ↗

Tsinghua backed humanoid robot startup raises $560 million and puts its robots on SF Express sorting lines

Star Dynasty, the only humanoid robot startup directly affiliated with Tsinghua University, has raised more than 4 billion yuan (about $560 million) in cumulative funding and deployed its bipedal humanoid robots into SF Express logistics sorting centers, per Pandaily. Founded in 2023 by Tsinghua assistant professor Chen Jianyu, the company pairs its STAR1 humanoid robot with an in-house dexterous hand and embodied AI brain to sort parcels at speeds matching experienced human workers. Unit deliveries grew 300% quarter over quarter after thousand-unit shipments began in the second quarter of 2026, with China Post and SF Express anchoring the rollout. Co-founder Xi Yue said the company chose logistics over automotive factory work because sorting still benefits from robots operating at 50 to 60% of human capability.

Read at Pandaily ↗

Alibaba upgrades its video generation model and launches a global AI filmmaking competition

Alibaba released HappyHorse 1.1, an upgrade of its video generation model, and paired the launch with a global AI filmmaking competition staged with Huajing Entertainment Group, per TechNode. The new version delivers improvements in motion dynamics, subject consistency, prompt adherence and visual and audio quality over the version 1.0 model released in April. The launch falls under the Cyberspace Administration of China's generative AI rules requiring synthetic media labeling and provider registration.

Read at TechNode ↗

ByteDance plans early-July Seedance 2.5 release, billed as foundation for world models

ByteDance's cloud arm Volcano Engine announced on June 23 that it will release Seedance 2.5, the next version of its video generation model, in early July, with the model in final internal testing, Caixin reported. Volcano Engine head Tan Dai told the publication that film and short drama are minor scenarios for Seedance. The model is being built as the foundation for world models, he said. Seedance 2.0, released in February, accelerated the AI conversion of China's short drama industry, with most short dramas on the market now generated by AI. Seedance is already subject to the Cyberspace Administration of China's generative AI provider registration and synthetic media labeling rules.

Read at Caixin ↗

III.Policy Tracker

House Science panel to mark up 10-bill AI package Thursday

The House Science, Space and Technology Committee is set to mark up 10 measures Thursday, June 25, nearly all targeting artificial intelligence, per the committee's notice. The slate is led by the CREATE AI Act, which would establish a National Artificial Intelligence Research Resource for compute and data access, and spans AI security and vulnerability reporting (AI Security and Innovation Act; AI Flaw Reporting and Security Enhancement Act), NIST standards and transparency (READ AI Models Act; Protecting Consumers from Deceptive AI Act; AI-Ready Federal Data Guidelines Act), NSF education and workforce (NSF AI Education Act; LIFT AI Act; Workforce for AI Trust Act) and data center energy use (Data Infrastructure Energy Measurement and Standards Act).

Read at House Science Committee ↗

IV.Industry & Market Watch

Midjourney unveils 'Midjourney Medical,' an underwater body scanner, as physicians urge caution

Midjourney, the AI image generator, has launched Midjourney Medical, an ultrasound scanner that lowers users into a tub of water and uses a ring of sensors to build a 3D body map in about 60 seconds, Business Insider and The Verge reported. The company billed the system as "as powerful as MRI, and as casual as a trip to the spa," and founder David Holz said Midjourney is not currently using AI in the device but hopes to later; the lab plans a San Francisco spa with 10 scanners opening by 2027 and 50,000 units worldwide by 2031. Radiologists told Business Insider that frequent full-body scanning can yield incidental findings, false positives, anxiety and unnecessary follow-up care without clinical context, and science communicator Hank Green said ultrasound does not replace MRI, CT or other scans. Midjourney, which has not finalized its data policies, said it will start with body-composition maps that avoid diagnostic clearance and submit results to the FDA for expanded capabilities.

Read at Business Insider ↗ • Read at The Verge ↗

Leaked Social Design Agency files reveal Russian plans for AI driven influence operations in Europe

Leaked files from Russia's Social Design Agency (SDA) reveal Kremlin-directed plans to seed AI training data and search results with state aligned content, Bloomberg reported. A "Project 2026" document lists an AI driven self-filling knowledge base aimed at Germany with servers launched and a database already containing over 200,000 pages, per OCCRP. The same document lists "AI News," a project for France that seeks to create three internet operations across six social networks producing hundreds of videos. The SDA has been sanctioned by the United States, the United Kingdom and the European Union, with the UK adding 49 SDA linked individuals to its sanctions list earlier this month.

Read at Bloomberg ↗ • Read at OCCRP ↗

Getty Images and OpenAI sign multiyear 'Display Partnership' to surface licensed images in ChatGPT

Getty Images announced a multiyear Display Partnership with OpenAI that will surface its licensed content inside ChatGPT's search and discovery, per Forbes and Engadget. CEO Craig Peters said high-quality, licensed visual content makes AI-powered search and discovery more useful and trustworthy. The arrangement is "display only," a Getty spokesperson told Engadget, with the announcement not specifying whether the images may be used to train OpenAI's models — a notable line for a company that banned AI-generated images in 2022 and sued Stability AI over alleged copyright infringement. Getty struck a similar display-only deal with Perplexity in October 2025; both OpenAI and Perplexity face suits alleging they scraped copyrighted material.

Read at Forbes ↗ • Read at Engadget ↗

WSJ details Altman-linked startup gains from OpenAI deals as House Oversight presses for documents

OpenAI's deals with Sam Altman linked startups including Helion Energy and Stoke Space have benefited Altman's personal investment portfolio, the Wall Street Journal reported Monday. Altman owns roughly one-third of Helion, a stake valued at approximately $1.65 billion as of late 2025, with financial disclosures showing he has more than $2 billion in companies that do business with OpenAI, per GeekWire. The House Oversight Committee sent Altman a letter on May 8 requesting documents related to OpenAI's handling of potential conflicts, citing the Helion relationship specifically. Altman stepped down from the Helion board in March 2026 to allow the companies to explore future partnerships.

Read at WSJ ↗