WH-Anthropic pivot to AI security vulnerability rules | FERC opens AI grid lane | $43M floods congressional AI fight

Bottom Line Up Front

The White House and Anthropic are jointly drafting a federal framework to grade the severity of AI security flaws, with Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross leading the administration side. The Federal Energy Regulatory Commission unanimously approved orders requiring grid operators to revise how data centers connect to the bulk power system, letting flexible large customers clear study in as little as 60 days. AI focused super PACs have spent $43.3 million on U.S. congressional races this cycle, with both pro-AI and restriction-minded groups targeting Senate, House and local contests, per OpenSecrets data cited by NPR. President Trump told Axios he no longer views Anthropic as a national security threat following direct talks with CEO Dario Amodei.

I.Top Stories

White House and Anthropic open joint talks on a federal framework for grading AI security flaws

Anthropic and senior administration officials are jointly developing a framework to assess the severity of security flaws in new AI models and guide future government intervention, per Politico, citing a senior White House official and a second administration official. The Anthropic side is led by head of public policy Sarah Heck and co-founder Tom Brown, with Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross representing the administration. The framework would create common benchmarks for assessing future jailbreaks, including the extent to which safeguards were bypassed, the capabilities exposed, and the practical consequences of the breach. Per Axios, Commerce officials are also slated to meet with Anthropic senior staffers, with separate sessions planned at the CIA and with White House science adviser Michael Kratsios.

Read at Politico ↗ • Read at Axios ↗

FERC unanimously approves large-load interconnection orders to speed AI data center grid hookups

The Federal Energy Regulatory Commission on Thursday issued a set of orders requiring U.S. interstate grid operators to justify or revise the rules governing how data centers and other large electricity customers connect to the bulk power system, NVIDIA wrote in a blog post welcoming the action. The orders follow Energy Secretary Chris Wright's directive instructing FERC to address large-load interconnection, and allow customers that can shift or curtail demand during peak hours to clear study in as little as 60 days. Large customers must fund their own network upgrades and bring new generation online alongside their demand. NVIDIA said it and Emerald AI are building AI factories designed as flexible grid assets, with commercial deployment beginning later this year.

Read at NVIDIA ↗ • Read at FERC ↗

AI focused super PACs have spent $43.3 million on congressional races this cycle, NPR reports

AI focused super PACs have already poured $43.3 million into U.S. congressional races for the current election cycle, NPR reported Monday, citing OpenSecrets data. The spending is hitting Senate races, House races and local offices, with both pro-AI and AI restriction camps mobilizing. NPR ties the surge to growing American discomfort with AI's impact on jobs and energy bills, citing local data center protests in Utah.

Read at NPR ↗

Trump says he no longer views Anthropic as a national security threat

President Trump told "The Axios Show" in an interview published Friday that he no longer considers Anthropic a U.S. national security threat. Asked whether he saw the company or CEO Dario Amodei as a threat, he said, "Well, not now, but a week ago, maybe." Trump credited Amodei with responding quickly and responsibly to the administration's export control directive, which last week barred foreign access to Anthropic's most advanced models, Mythos 5 and Fable 5. He did not rule out invoking the Defense Production Act against the company.

Read at Axios ↗ • Read at Reuters ↗

Vance stakes out a distinct MAGA vision for AI

Vice President JD Vance has become a leading administration voice on AI, pairing a hands-off stance on regulation with a warning against letting a few dominant firms control the technology, The Atlantic reported. Vance said he has "not yet seen the evidence" that AI will cause widespread job loss but worries its gains will fall disproportionately on the wealthy. He favors a single national standard over state-by-state rules, calling it the "worst possible outcome" to let California set the regulatory map, and has invoked Pope Leo XIV in arguing that "decisions over life and death must be made by humans and not machines."

Read at The Atlantic ↗

II.China Watch

Eight Chinese ministries roll out a national plan to push AI into consumer products, retail and elderly care

China's Ministry of Commerce and seven other ministries, including the Central Cyberspace Affairs Commission Office, the National Development and Reform Commission and the Ministry of Industry and Information Technology, jointly issued "Implementation Opinions on Accelerating the Development of AI+Consumption," Sinocism reported. The guideline sets out 17 measures across five areas, including AI powered robots for elderly care and household assistance, smart wearables, AI integrated e-commerce and "AI+Consumption" experience centers. Local governments are authorized to subsidize next generation smart terminals under the consumer goods trade-in policy framework. The rollout extends Beijing's broader "AI+" initiative from industry into the consumer economy. It gives the Cyberspace Administration of China and sector regulators a fresh policy hook for product level AI rules, content review and registration regimes as agentic and embodied systems move into Chinese households.

Read at Sinocism ↗

Beijing puts 10 U.S. drone, robotics and rare earth firms on dual-use export control list, citing Pentagon list expansion

China's Ministry of Commerce on June 22 added 10 U.S. entities to its export control list under China's Export Control Law and Regulations on the Export Control of Dual-Use Items, Caixin reported. The targets include MP Materials and USA Rare Earth, the two domestic rare earth producers Washington has backed to reduce reliance on China, along with drone and robotics makers Red Cat Holdings, Teal Drones and Jaia Robotics. Chinese exporters are barred from shipping dual-use items to those entities, any party worldwide is prohibited from transferring China-origin dual-use goods to them, and ongoing exports must stop immediately. MOFCOM said the action responds to Washington's recent expansion of the Section 1260H list of Chinese military companies, with the new additions spanning defense, aerospace, robotics, rare earths and drones. Exporters facing genuine necessity must apply to MOFCOM for case-by-case approval. The retaliation uses the same dual-use legal framework Beijing has built out for AI chip, model and software export licensing, and squeezes rare earth and drone component flows that feed U.S. AI-enabled autonomous systems.

Read at Caixin ↗ • Read at MOFCOM ↗

Zhipu's Hong Kong cap crosses HK$1 trillion as Chinese AI flagship's stock gains roughly 1,900% since IPO

Zhipu AI's Hong Kong shares jumped more than 13% on June 22 to push market capitalization above HK$1 trillion ($130 billion), up roughly 1,900% since its IPO less than six months ago, Pandaily reported. AIPD's June 18 edition covered the prior milestone when shares lifted the cap past HK$930 billion ($120 billion) alongside the open source GLM-5.2 release. Zhipu's 2025 annual report showed 724 million yuan ($110 million) in revenue against a 4.7 billion yuan ($690 million) net loss, with only about 4% of shares freely tradable and a July 8 lockup expiration covering roughly 25.7 million shares. GLM-5.2 ranks in the top two of the Code Arena blind benchmark and recently topped the Design Arena leaderboard, with the Coding Plan subscription selling out daily quotas within seconds. The HK$1 trillion crossing coincides with Hang Seng Tech Index inclusion effective June 8 and tightened U.S. frontier model export controls.

Read at Pandaily ↗

Moore Threads' 2025 revenue tripled as China's domestic GPU substitution accelerates

Chinese GPU maker Moore Threads reported 1.505 billion yuan ($220 million) in 2025 revenue, up 243% year over year, with Q1 2026 revenue continuing the trajectory, Pandaily reported. The company credited the surge to rising Chinese AI demand and U.S. export restrictions on advanced GPUs, which opened a substitution window for domestic alternatives. Commercial deployment of the flagship MTT S5000 GPU and the KUAE 10,000-card intelligent computing cluster expanded revenue, with a 660 million yuan ($98 million) cluster contract signed in March 2026. Moore Threads has achieved Day-0 adaptation for DeepSeek V4 and major Chinese open source models, with its MUSA developer ecosystem now exceeding 450,000 users. The growth bolsters Beijing's domestic AI compute base under U.S. Commerce Department export controls and gives Chinese AI developers a substitute for sanctioned U.S. accelerators in training and inference workloads.

Read at Pandaily ↗

III.Policy Tracker

Commerce Secretary Lutnick pressed ASML in recent meetings on suspected EUV machine in China

Commerce Secretary Howard Lutnick told senior ASML executives in a series of recent meetings that he was concerned a Dutch extreme ultraviolet lithography machine may have ended up in China, per Bloomberg. Senior administration officials told Bloomberg they have evidence ASML shipped EUV related components and transport equipment to China but have declined to share that evidence. ASML denies that any EUV machine exists in China and says one never has. EUV systems are the only tools capable of printing the most advanced semiconductor patterns, and selling EUV to China has been banned since the first Trump administration.

Read at Bloomberg ↗ • Read at TechCrunch ↗

VA IT official tells contractors AI delivery will drive mid-term reviews, incumbency offers no protection

Zack Schwartz, principal deputy assistant secretary in the Department of Veterans Affairs Office of Information and Technology, told contractors following a private Advanced Planning Brief to Industry on June 17 that "incumbency is not a guarantee, incumbency is not an advantage" on AI delivery, FedScoop reported. Schwartz said the VA will review existing contracts mid-term against modernization and AI expectations and is open to replacing vendors that fall behind. He credited Secretary Doug Collins for backing the shift, and said the VA, the fourth-largest spender in the U.S. government, has obligated $177.7 billion so far this fiscal year. Schwartz said AI is becoming an expectation across every VA procurement.

Read at FedScoop ↗

Waymo files NHTSA recall covering more than 3,800 robotaxis after construction-zone incursions

Alphabet's Waymo filed a voluntary software recall with the National Highway Traffic Safety Administration covering more than 3,800 fifth-generation robotaxis after at least 13 incidents in which vehicles drove past ramp closure signs into preplanned freeway construction zones, per the Wall Street Journal. Six of the incidents occurred in Phoenix in April and seven in the San Francisco Bay Area in May, per TechCrunch. Waymo restricted all highway operations on May 19 and is rolling out a software fix; the company's Field Safety Committee approved the recall on June 8. It is Waymo's sixth recall and follows a May recall after vehicles drove into flooded roads, plus an open NHTSA investigation into a January incident in which a Waymo robotaxi struck a child near a Santa Monica elementary school.

Read at WSJ ↗ • Read at TechCrunch ↗

IV.Capability & Research Watch

Google DeepMind publishes AI Control Roadmap v0.1, introducing TRAIT&R adversary taxonomy and two defensive invariants

Google DeepMind on June 18 published the GDM AI Control Roadmap (v0.1), its plan for internal guardrails meant to catch adversarial behavior by AI agents even as they get harder to oversee, per the Alignment Forum. The document introduces TRAIT&R, a taxonomy of tactics and techniques available to a hypothetical AI adversary modeled on the MITRE ATT&CK security framework, and sorts hypothetical threats into loss of control, work sabotage and direct harm. It establishes two defensive invariants meant to hold as adversary capability grows: reliable detection of misaligned intent or actions, and effective prevention and response. The roadmap outlines four detection tiers and three response tiers and proposes 15 concrete mitigations ranging from chain-of-thought monitoring to system-level anomaly detection and shutdown infrastructure.

Read at Google DeepMind ↗ • Read at Alignment Forum ↗

LessWrong essay urges intermediate biorisk warning levels after Anthropic activated Mythos protections without crossing CB-2

Anthropic's Mythos and Fable 5 system card states the model meets the CB-1 (non-novel) biological and chemical weapons capability threshold but falls short of CB-2 (novel), yet activates protective measures in response to both, per a LessWrong essay published Sunday. The author argues that growing gaps between threshold-triggered and actual governance decisions show that real safety calls are being made in what Anthropic's Responsible Scaling Policy v3 itself calls a "zone of ambiguity." The piece proposes intermediate warning levels between terminal biorisk thresholds to govern that gray area, and cites a 2025 precedent in which Anthropic activated ASL-3 protections with Claude Opus 4 despite uncertainty over whether thresholds had been met.

Read at LessWrong ↗ • Read at Anthropic ↗

V.Industry & Market Watch

Humans First plans 13-location July 18 protest against AI data centers across five states

Humans First, a conservative organization that bills itself as an "America First AI policy" group, told Axios it is organizing a Nationwide Day of Protest against what it calls unchecked expansion of AI data centers on July 18, with rallies set in at least 13 locations across Georgia, California, Texas, Florida and Virginia. Group chair Amy Kremer, a longtime Tea Party figure who helped organize the rally preceding the Jan. 6, 2021 attack on the Capitol, said in a statement that the "disconnect between the elites and the base that gave rise to the Tea Party movement can be seen today in the battle over AI data centers." The group's rally site lists secrecy, water use, air pollution, national security, energy, land use and noise as data center harms, and Kremer said the protests will give grassroots conservatives a voice on data center policy. Humans First told Axios it welcomes liberal participation though its primary focus is rallying conservatives.

Read at Axios ↗

Anthropic warned about advanced AI dangers far more than OpenAI through 2026, FT analysis finds

Anthropic publicly warned about the dangers of advanced AI substantially more often than OpenAI through 2026, a Financial Times analysis published Saturday found. Per TechCrunch, OpenAI CEO Sam Altman had publicly cast Anthropic's framing of Mythos in April as "fear-based marketing," accusing the company of building a perceived threat and selling expensive safeguards against it. More than 100 cybersecurity experts from companies including Adobe and Nvidia signed an open letter June 14 asking the administration to lift its directive restricting export of Anthropic's Mythos models, arguing the models are good but not uniquely good at finding software vulnerabilities and that pulling them weakens U.S. defenders against advancing Chinese capability, per AP.

Read at FT ↗ • Read at TechCrunch ↗ • Read at AP ↗

Chevron and Microsoft sign 20-year natural gas power deal for proposed West Texas data center

Chevron has signed a 20-year exclusive deal with Microsoft to supply natural-gas-fired power for a proposed West Texas data center that could be one of the biggest in the U.S., Bloomberg reported. The deal is tied to a roughly $7 billion gas-fired plant Chevron is building with investment fund Engine No. 1, with an initial 2,500-megawatt nameplate intended to power a large data center campus, per StreetInsider citing the prior Reuters disclosure. Chevron had previously said its first AI data center gas project would target a West Texas start-up by 2027.

Read at Bloomberg ↗ • Read at StreetInsider ↗

John Jumper to leave Google DeepMind for Anthropic after nearly nine years

John Jumper said Friday he will leave Google DeepMind to join Anthropic after nearly nine years, CNBC reported. Jumper shared a 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for AlphaFold, the AI system that has predicted more than 200 million protein structures. The move follows Gemini co-lead Noam Shazeer's earlier announcement that he is leaving DeepMind for OpenAI.

Read at CNBC ↗

VI.Global & Geopolitics

Norway imposes near-ban on generative AI for junior school pupils, restricts use for older children

Norway's prime minister has imposed a near-ban on the use of generative AI tools by junior school pupils and restricted their use in the education of older children, citing the risk that AI use lets young children skip foundational learning, per the South China Morning Post. The government framed the move against a broad decline in education test scores. The measure follows Norway's 2024 ban on smartphones in schools and an expansion of teacher disciplinary powers.

Read at SCMP ↗

Microsoft attributes supply chain attack to North Korea's BlueNoroff

Microsoft has attributed the recent compromise of more than 140 npm packages in the Mastra AI developer ecosystem to Sapphire Sleet, the North Korean state group also known as BlueNoroff that primarily targets the financial sector, BleepingComputer reported. The attackers compromised the npm maintainer account "ehindero" and used it to publish malicious updates across more than 140 @mastra packages. One of those updates was a typosquat dependency called "easy-day-js" masquerading as the legitimate dayjs JavaScript library.

Read at BleepingComputer ↗