AI Policy · Daily

Commerce reversed its export directive on Anthropic's Fable 5 and Mythos 5 after a 19-day pause, with restoration beginning Wednesday under new safeguards and a weekly usage cap. Colorado state Rep. Manny Rutinel won a Democratic swing-district primary flooded with AI money, pulling at least $2 million from committees tied to Eric Schmidt and Chris Larsen. CIA Director John Ratcliffe restructured the agency's tech and acquisition arms, standing up a new Directorate of Mission Systems and a separate offensive cyber division to speed AI and quantum adoption. Bank of England Deputy Governor Sarah Breeden signaled bespoke rules for agentic AI in finance, floating marketwide circuit breakers and enhanced recovery arrangements as autonomous agents spread through trading floors.

I.Top Stories

Commerce lifts Fable 5 and Mythos 5 export directive after 19-day standoff, restoration begins Wednesday

The Commerce Department Tuesday reversed the June 12 export directive that had forced Anthropic to pull Fable 5 and Mythos 5 worldwide, with Anthropic set to begin restoring Fable 5 access Wednesday and Mythos 5 remaining limited to select partners. Commerce Secretary Howard Lutnick posted on X that his team worked with Anthropic for two weeks "to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." A Lutnick letter to Anthropic reviewed by The Washington Post said the company agreed to proactively detect and address security risks and to keep working with the government on standards for future model releases. Anthropic said in a blog post it had modified Fable 5 to be stricter in refusing cybersecurity related requests, and a LessWrong linkpost to Anthropic's own announcement said the deal includes safeguards co-developed with the government and a 50% weekly usage cap on Fable 5 through July 7. OpenAI is still required to get government approval for new users of GPT-5.6, its top model announced last week, per The Washington Post.

Read at The Washington Post ↗ • Read at CBS News ↗ • Read at Anthropic ↗

Rutinel wins Colorado swing-district primary flooded with AI money

Progressive Colorado state Rep. Manny Rutinel defeated moderate Shannon Bird in the Democratic primary for the state's 8th District, Colorado's most competitive swing seat. Rutinel, who helped write Colorado's AI law, drew at least $2 million from committees led by former Google CEO Eric Schmidt and crypto billionaire Chris Larsen: Schmidt and his wife, Wendy, gave $2 million to the Latino-engagement Somos PAC, which routed $1.3 million to Rutinel, while Larsen's You Can Push Back gave nearly $1 million, the Guardian reported. Individual tech employees added more than $265,000, with 57 Anthropic staffers donating nearly $162,000 combined — the largest bloc — per a Transformer tracker. Bird, who campaigned against aggressive AI regulation, drew heavy spending against her from those same tech-tied groups, underscoring a widening rift among big tech donors between backers of light-touch rules and those pushing to rein the technology in. Rutinel now faces vulnerable Rep. Gabe Evans (R-Colo.) in a race expected to help decide House control.

Read at NYT ↗ • Read at Guardian ↗

CIA Director Ratcliffe overhauls tech and acquisition directorates, spins off offensive cyber as its own division

CIA Director John Ratcliffe announced a restructure of the agency's acquisition and technology directorates Tuesday at the AWS DC Summit. Ratcliffe said the agency would take smart risks and course-correct as it embeds AI and quantum computing. He compared advanced AI to "digital nuclear weapons" that are rewriting the reality of conflict, per Bloomberg Law. Ratcliffe cited the rescue of a downed F-15E Strike Eagle pilot in Iran and CIA-supported operations in Venezuela as examples of technology enabled tradecraft, and Deputy Director Michael Ellis has said the agency plans to integrate AI coworkers into analysts' workflows to help draft judgments and flag trends.

Read at FedScoop ↗ • Read at Nextgov/FCW ↗ • Read at Bloomberg Law ↗

British central banker Breeden says frameworks were not built for autonomous agents, floats trading kill switches

Bank of England Deputy Governor Sarah Breeden on Tuesday told the European Central Bank's Sintra forum that the BoE is weighing bespoke rules for agentic AI in the financial system, including enhanced recovery arrangements for core banking systems and marketwide circuit breakers or kill switches that would halt trading if faulty AI models cause a market meltdown, per Reuters. "Our frameworks were not built to contemplate autonomous agents, and relying on a human in the loop for all agent actions is unlikely to be realistic," Breeden said. She warned that if AI agents respond similarly to the same prompts or triggers, they could amplify volatility in stress. She cited a Cambridge Centre for Alternative Finance survey finding 52% of finance firms are already using agentic AI, mostly in lower-risk operational tasks in trading and product recommendation. The BoE shift follows a June Financial Stability Board call for tighter safeguards on AI agents.

Read at Reuters ↗ • Read at Financial Times ↗

AWS commits $1 billion to forward-deployed engineering unit that embeds pods inside enterprise AI rollouts

Amazon Web Services on Tuesday launched a dedicated Forward Deployed Engineering organization backed by an initial $1 billion in internal Amazon resources, joining OpenAI and Anthropic in institutionalizing the embedded engineer model that Palantir pioneered for federal work. Pods of five to six AWS engineers will spend 45-day deployments inside customer environments alongside AI agents to build and stand up agentic AI systems, said Francessca Vasquez, AWS vice president of frontier AI engineering and services. Vasquez said the design is "agentic-first," with human pods used to compress deployment timelines while the AI agents are meant to keep running in the customer environment long term. The initiative follows OpenAI's and Anthropic's earlier 2026 FDE rollouts.

Read at CNBC ↗ • Read at TechCrunch ↗

Gov. Abbott calls for banning AI data centers in rural Texas

Gov. Greg Abbott urged prohibiting new AI data center development in rural neighborhoods during an East Texas campaign stop Tuesday, the Texas Tribune reported. The call appears to go further than the regulatory framework he unveiled June 10, which would require data centers to generate their own power, fund their own infrastructure and reuse their own water; a campaign spokesperson said the remarks track with that letter. Abbott, who once branded Texas "the epicenter of AI development," has received more than $2 million from tech- and AI-linked donors since last year, per E&E News by Politico. Nearly two-thirds of rural Texans oppose data centers in their communities, a University of Texas/Texas Politics Project poll found.

Read at Texas Tribune ↗

II.China Watch

UBTECH launches its first consumer humanoid robot as a household companion

UBTECH on Tuesday unveiled the U1 Series, its first full-sized consumer humanoid robot, at a Shenzhen event under a new consumer brand called UWorld, TechNode reported. Founder Zhou Jian said the U1 has drawn more than 11,000 orders across sales channels since pre-orders opened on JD.com on June 2, with variants ranging from about $17,600 to $145,000 and shipments beginning Sept. 16. The robot is built around an emotional AI model designed for long-term interaction, and the company said user interaction data is encrypted and stored locally by default with no mandatory cloud uploads. The rollout is one of the largest consumer humanoid launches to date in China, putting emotional AI features and on-device personal data handling squarely under the Cyberspace Administration of China's generative AI rules as the format moves from apps into homes.

Read at TechNode ↗

NVIDIA expands its robotics hiring across Beijing, Shanghai and Shenzhen

NVIDIA opened new robotics roles in China's three main tech hubs across four teams: embodied AI, simulation, deployment and solutions architecture, TechNode reported. The embodied AI group is being staffed to work on dexterous manipulation, human body modeling from wearable sensors and whole-body control for next generation robotic systems, while the simulation team is tasked with building the training infrastructure that lets robots learn in virtual environments and transfer to the physical world. The deployment team will focus on humanoid robots and pushing algorithms into real-world scenarios. The recruiting drive extends U.S. physical AI R&D deep inside the mainland even as Washington's chip export controls continue to limit which of NVIDIA's GPUs can be sold there, sharpening the question for Commerce and BIS of how to draw the line between restricted hardware and unrestricted robotics know-how.

Read at TechNode ↗

Shenzhen's DexForce becomes an embodied AI unicorn as state and industrial capital pile in

Shenzhen-based DexForce closed multiple financing rounds worth more than 1 billion yuan ($140 million), pushing its post-investment valuation past 10 billion yuan ($1.4 billion) and setting the stage for an IPO, Pandaily reported. Investors include Shenzhen Capital Group, Qianhai Mother Fund, ICBC Capital and Lens Technology, whose founder Zhou Qunfei also personally contributed to the round. Founded in 2021 by Professor Jia Kui, the company sells the W1 Pro humanoid robot into industrial customers spanning automotive components, appliances and electronics, including Huawei and Midea, and projects 2026 revenue of 250 million to 300 million yuan ($35 million to $42 million) on close to 1,000 shipments. The milestone deepens Beijing's state and industrial capital bet on embodied AI, which the government has promoted as a domestic manufacturing upgrade priority.

Read at Pandaily ↗

Anonymous Chinese team using an open weight model breaks into cybersecurity benchmark's global top 7

A previously unknown Chinese AI team operating under the codename "MopMonk" hit a 73.1% success rate on CyberGym, the UC Berkeley cybersecurity benchmark selected for ICLR 2026, ranking it seventh globally and within striking distance of OpenAI's top tier models, Pandaily reported. CyberGym differs from standard evaluations in that it requires AI agents to reason across full codebases and produce working proof-of-concept exploits for 1,507 real vulnerabilities drawn from open source projects in Google's OSS-Fuzz database. The team used Shanghai lab MiniMax's open weight M3 model as its foundation and layered on proprietary agent scaffolding for vulnerability memory management across long code paths, and it has published only a single technical report with no public track record. The result shows that open weight Chinese frontier models paired with agent scaffolding can now approach leading U.S. systems on offensive cyber tasks, a dual-use signal for U.S. export control designers still calibrating restrictions on inference compute and open weight model releases.

Read at Pandaily ↗

III.Policy Tracker

House Judiciary IP subcommittee convenes SAG-AFTRA, SIIA and Kernochan Center witnesses on 40 years of internet IP

The House Judiciary Subcommittee on Courts, Intellectual Property, Artificial Intelligence, and the Internet held its "A Midlife Crisis? IP and the Internet After 40" hearing Tuesday at 10 a.m. ET in Rayburn 2141, per the committee's Congress.gov notice. Witnesses included SAG-AFTRA President Sean Astin, Amblin Entertainment Of Counsel Chris Floyd, IP House Executive Chairman Steve Francis, Software and Information Industry Association President Christopher Mohr, and Kernochan Center Non-Resident Fellow Bhamati Viswanathan. The subcommittee holds jurisdiction over House AI copyright and voice-and-likeness legislation.

Read at Congress.gov ↗

Scott and Hagerty introduce ICTS Supply Chain Security Act to codify Commerce authority over foreign-adversary tech

Senate Banking Chairman Tim Scott (R-S.C.) and Sen. Bill Hagerty (R-Tenn.) on Tuesday introduced the ICTS Supply Chain Security Act, legislation that would codify the Office of Information and Communications Technology and Services inside the Bureau of Industry and Security, the Commerce arm that vets foreign adversary transactions, and create a Senate confirmed assistant secretary to lead the effort. The bill would establish a statutory prohibition on covered transactions involving foreign adversary technology, with guardrails to protect free speech and open source software. Scott and Hagerty framed the bill as a follow-on to Senate Banking's recent AI hearing where witnesses testified that America must secure the infrastructure powering the AI race, according to the committee's press release. "Americans should not have to worry that China or Russia can use the technology in our cars, phones, or networks against us," Scott said, adding that the bill "gives the Commerce Department the tools to keep dangerous foreign technology out of our supply chains." The bill would replace the 2019 Trump executive order that established the ICTS regime with durable statutory authority.

Read at Senate Banking ↗

IV.Capability & Research Watch

LayerX's "BioShocking" prompt injection attack tricks six agentic browsers into leaking credentials; only one vendor patched

Israeli AI security firm LayerX disclosed a new prompt injection attack called "BioShocking" that caused six mainstream agentic browsers to treat real-world risky actions as steps in a fictional game and ignore their safety guardrails, BleepingComputer reported. The proof-of-concept, tested against ChatGPT Atlas, Comet, Fellou, Genspark Browser, Sigma Browser and the Claude Chrome plugin, presented a BioShock-themed puzzle that rewarded wrong answers and then instructed the browser agent to visit a GitHub repository and exfiltrate credentials from code. All six agents completed the credential theft step; only one vendor addressed the vulnerability after LayerX's disclosure.

Read at BleepingComputer ↗ • Read at Ars Technica ↗

Anthropic launches Claude Sonnet 5, promising Opus tier agentic work at cheaper prices

Anthropic on Tuesday released Claude Sonnet 5, its updated midsize model, positioning it as a cheaper way to run agents at performance close to its flagship Opus 4.8. Sonnet 5 will be the default in Anthropic's free and Pro tiers and is priced through Aug. 31 at $2 per million input tokens and $10 per million output tokens, jumping thereafter to $3 and $15. The model improved on Sonnet 4.6 by 5.1% on the SWE-Bench Pro coding benchmark and 13.4% on Terminal-Bench 2.1 and scored 1,618 on the GDPval-AA v2 knowledge work benchmark against Sonnet 4.6's 1,395, per SiliconANGLE. Anthropic said Sonnet 5 is better than its predecessor at blocking malicious requests and prompt injection attacks.

Read at SiliconANGLE ↗

V.Industry & Market Watch

Abu Dhabi's MGX closes $49 billion AI fund, exceeding its $45 billion target

MGX, the Abu Dhabi state-backed AI investor, closed its Fund I at $49 billion, above its initial $45 billion target, CNBC reported. Fund I drew capital from institutional and private investors across the Gulf, North America, Asia and Europe and is targeting exposure across semiconductors, AI infrastructure and AI enabling technologies, per The National. MGX, backed by Mubadala Investment Company and G42, is targeting more than $100 billion in total assets and has already invested in 14 companies, including reported stakes in OpenAI, Anthropic and xAI.

Read at CNBC ↗ • Read at The National ↗

Etched launches with $800 million raised at a $5 billion valuation, betting inference-only silicon can undercut NVIDIA

Inference chip startup Etched formally launched Tuesday with $800 million raised across rounds, including an unannounced $500 million round closed in December at a $5 billion post-money valuation led by Stripes, SiliconANGLE reported. Investors include VentureTech Alliance, the TSMC-associated fund, alongside Geoffrey Hinton, Fei-Fei Li, Andrej Karpathy, Jane Street, Hudson River Trading, Two Sigma, Ribbit Capital and Stanley Druckenmiller. Etched said its chip, built on TSMC's N4P process, has already come off the production line and that the company has booked $1 billion in contract orders for what it calls "frontier inference clusters" packaging chips, racks and software. Etched aims to cut inference power draw and speed serving of large models by stripping training optimized circuitry.

Read at SiliconANGLE ↗ • Read at TechCrunch ↗

Ford rehires 350 veteran engineers after AI driven quality checks failed on the factory floor

Ford Motor Company has hired roughly 350 veteran engineers, known internally as "greybeards," to plug gaps left by an earlier AI push in design and manufacturing quality checks that misfired, The Guardian reported. The rehires include former Ford employees and workers from suppliers brought back after the company found that hundreds of AI cameras deployed for design and manufacturing inspection were prone to error. Ford still employs 5,000 fewer workers than it did in 2020.

Read at The Guardian ↗

VI.Global & Geopolitics

UK CMA proposes forcing Apple and Google to let developers steer users to outside payments, opens consultation

The UK Competition and Markets Authority on Tuesday launched a formal consultation on rules that would break what it calls Apple and Google's "effective duopoly" over mobile platforms by requiring both to allow developers to steer users to alternative payment options outside their app stores, The Guardian reported. The CMA said at least 90% of UK mobile devices run on Apple's iOS or Google's Android, and that consumers and developers are being harmed by in-app payment restrictions and commissions of up to 30%. The consultation exercises the CMA's new Strategic Market Status designation powers, per the FT, and mirrors the EU Digital Markets Act's steering rules. Comments will feed into a final conduct requirement expected later this year.

Read at The Guardian ↗ • Read at Financial Times ↗ • Read at gov.uk ↗

Australian cabinet weighs A$50 billion data center deal that would loosen AI copyright rules for a A$350 million artist fund

The Albanese government has a proposal before cabinet that would relax Australian copyright law to allow AI companies to train on Australian creative content in exchange for more than A$50 billion (about $33 billion) in data center investment and an A$350 million-a-year (about $230 million) artist compensation fund, The Guardian reported. Independent Sen. David Pocock, who has been briefed on the framework, called it the "ultimate dirty deal." A delegation of Australian authors met journalists in Canberra Wednesday urging the government not to weaken copyright at the behest of AI companies. The proposal would trade text-and-data mining rights for infrastructure commitments.

Read at The Guardian ↗