--- date: 2026-07-17 subject: "'AI safety' defined | CAISI needs dedicated funding | BIS 'paralyzed' under Kessler" --- **POLITICO surveyed 20 industry figures on what "AI safety" means**, including AIPN's Mark Beall, and separated AI policy into seven factions. Transformer investigates the **need for dedicated funding for CAISI**. House China hawks **Moolenaar and Whitesides pressed the Trump administration** to bar U.S. purchases of some Chinese memory chips. **Commerce's BIS is 'paralyzed' under Jeff Kessler**, with slow chip export license decisions and muddled policy causing backlogs. **Xi Jinping used his first in-person WAIC keynote** to frame China's AI push as open but "secure and controllable," pressing Beijing's global governance agenda against Washington's export controls. # 1. AI Policy Today - **Politico maps the AI safety debate through 20 industry voices, including AIPN's Mark Beall** — Politico Magazine asked 20 prominent figures working in and around AI the same question — what "AI safety" means to them — and found in a feature published Friday that the term has become a kind of Rorschach test, signaling everything from existential risk to child protection to mass job loss, alongside skeptics who treat "safety" as a pretext to block technological progress. The magazine reported the divide doesn't track partisan or ideological lines, with skeptics and boosters on both the left and right; respondents included Sen. Bernie Sanders (I-Vt.), MIT physicist Max Tegmark, philosopher Nick Bostrom and Dean Ball, among others. AI Policy Network's Mark Beall defined AI safety as protecting U.S. national security from advanced AI threats like weaponization and loss of control, and building the technical capacity to keep such systems "under human stewardship." [POLITICO](https://www.politico.com/news/magazine/2026/07/17/ai-safety-regulation-debate-factions-00990453) - **CAISI should be legislated out from under NIST with dedicated funding, Transformer News outlines** — Transformer News on Thursday highlighted the need for legislating the Center for AI Standards and Innovation (CAISI) out from under the National Institute of Standards and Technology and giving it dedicated funding and independent authority, citing two sources familiar with the agency's role. CAISI was "only peripherally involved" in decisions to impose export controls on Anthropic's Mythos and Fable models and to delay then approve OpenAI's GPT 5.6, the essay said. Those calls were made instead by former AI czar David Sacks, Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick to a lesser extent. CAISI lead Chris Fall was briefed on Mythos and Fable's technical capabilities and his office tested the model days before it was taken offline, but those evaluations had little influence over what happened. [Transformer News](https://www.transformernews.ai/p/caisi-us-ai-agency-governance) - **Senate blocks Democratic CRA to end Medicare AI prior authorization pilot on 46-50 party-line vote** — Senate Republicans on Thursday blocked a Democratic resolution to halt a Medicare pilot that uses AI to approve or deny physician-ordered care in original Medicare, per The Hill. Sen. Ron Wyden (D-Ore.), ranking member of the Senate Finance Committee, sponsored the Congressional Review Act resolution (S.J.Res.198), which failed 46-50 along party lines, per Bloomberg Government. The pilot is run through the Centers for Medicare and Medicaid Services. It would apply prior authorization, rarely allowed in traditional Medicare, to physician-ordered care. CRA resolutions require only a simple majority to pass, so Democrats cannot unwind the AI adjudication model without Republican defections. [The Hill](https://thehill.com/policy/healthcare/5972792-senate-blocks-medicare-ai-pilot/) [Bloomberg Government](https://news.bgov.com/bloomberg-government-news/senate-democrats-attempt-to-block-medicare-ai-prior-auth-model) [CMS](https://www.cms.gov/newsroom/press-releases/cms-launches-new-model-target-wasteful-inappropriate-services-original-medicare) - **House China hawks urge Trump administration to bar U.S. companies from buying memory chips from CXMT and YMTC** — House China Committee Chair John Moolenaar and Rep. George Whitesides, D-Calif., urged the Trump administration to bar U.S. companies from buying memory chips from Chinese suppliers ChangXin Memory Technologies (CXMT) and Yangtze Memory Technologies (YMTC), per the Financial Times. The push targets Apple, which is lobbying Commerce for clearance to source DRAM from CXMT for the Chinese market during a global memory chip shortage, per CryptoBriefing. In a letter to Commerce Secretary Howard Lutnick, the lawmakers said every purchase from the firms would subsidize the People's Liberation Army, noting both appear on the Pentagon's 1260H list of Chinese enterprises with military ties. The Trump administration in June 2026 postponed adding CXMT to the Commerce Entity List during trade negotiations with Beijing. CXMT is set to complete China's largest domestic IPO of 2026 on July 27. [Financial Times](https://www.ft.com/content/9e7cdf3c-2e52-492f-afeb-b63d86a53ce6?syn-25a6b1a6=1) [The Register](https://www.theregister.com/systems/2026/07/17/chinese-memory-ban-would-cut-off-rampocalypse-relief/5273993) - **Bureau of Industry and Security 'paralyzed' under Kessler as chip export approvals slow** — The Bureau of Industry and Security, the Commerce Department office that clears every advanced U.S. chip export license, is "paralyzed" under Jeff Kessler as approval decisions slow, NBC News reported. More than a dozen government and industry insiders described "slow decision-making, muddled policy decisions and close embrace of industry partners" at the bureau. [NBC News](https://www.nbcnews.com/tech/tech-news/trump-export-lutnick-kessler-us-technology-bis-ai-control-security-rcna353749) - **Xi delivers first WAIC keynote, framing China's AI push as 'open' but 'secure and controllable'** — President Xi Jinping on Friday delivered the keynote at the opening of the World Artificial Intelligence Conference in Shanghai, his first in-person appearance at the summit since it launched in 2018, per SCMP. Xi said China would seize the "rare historic opportunity" of AI driven growth by "encouraging open source, openness, collaboration and sharing." He called for a people-centered approach that shares AI capacity with developing countries to prevent "new historical injustices." He paired the openness pitch with a call to keep the technology "secure and controllable," a formulation that recurs in Chinese AI governance drafts. Xi's personal appearance turns WAIC into a state level platform for Beijing to press its Global AI Governance Initiative and proposed World AI Cooperation Organization against Washington's export control agenda. [SCMP](https://www.scmp.com/tech/tech-war/article/3360870/top-takeaways-xi-jinpings-opening-address-world-ai-conference-shanghai) # 2. China Watch - **Beijing denounces three U.S. chip bills headed for NDAA as 'arbitrary disruption' of supply chains** — The Chinese embassy in Washington said Beijing rejects attaching the AI OVERWATCH Act, the MATCH Act and the Chip Security Act to the U.S. National Defense Authorization Act, arguing that "arbitrary disruption and damage to the global industrial and supply chains serves no one's interest," per SCMP. The statement is Beijing's first formal reaction to the Senate manager's amendment covered in AIPD's [July 15th edition](https://aipolicydaily.org/archive/daily/2026-07-15/), which would fold the AI OVERWATCH Act's export licensing regime for advanced AI chips, the MATCH Act's push on allied lithography sales and the Chip Security Act's chip location tracking requirement into the annual defense bill. The statement coincides with the World AI Conference in Shanghai, where Chinese officials are promoting an alternative "open" AI governance narrative. If enacted, the bundle would fix AI chip export rules in statute rather than in Commerce Department regulation, raising the bar for any future U.S. administration seeking to negotiate the controls back down. [SCMP](https://www.scmp.com/news/us/article/3360846/beijing-denounces-us-chip-curbs-threat-global-supply-chains) - **DeepSeek's first outside round pins the AI lab at $52 billion via a luggage maker filing** — Anhui Korrun Co., a domestic luggage manufacturer, announced Thursday that its subsidiary Ningbo Purun holds a 0.0114% indirect stake in DeepSeek from an investment implying a post-money valuation of about 350.88 billion yuan ($52 billion), per Caixin. The Hangzhou-based lab's first outside round raised roughly 50 billion yuan (about $7 billion) against a 300 billion yuan (about $42 billion) pre-money valuation, with proceeds earmarked for team incentives and technical breakthroughs. Recent investment rounds show Chinese frontier labs financing their compute buildouts entirely from domestic capital as U.S. outbound investment rules keep American money out. [Caixin](https://www.caixin.com/2026-07-17/102465129.html) - **Zhipu says its annualized revenue hit $1 billion after a five month coding push** — Chinese frontier lab Zhipu AI has reached $1 billion in annual recurring revenue as of July, up 15-fold since January, according to sources familiar with the company cited by 36Kr. Zhipu went from $100 million to $1 billion in about five months, compared with 15 months at Anthropic. Zhipu began concentrating its model research on coding in early 2025, with founder Tang Jie framing "Coding and Reasoning" as the capabilities that pair with agents. The milestone comes days after Zhipu's HK$31.4 billion (about $4 billion) H-share placement covered in AIPD's [July 13th edition](https://aipolicydaily.org/archive/daily/2026-07-13/). It gives Beijing a domestic frontier lab whose revenue base can now fund independent chip purchases even as U.S. export controls tighten access to advanced accelerators. [36Kr](https://www.36kr.com/p/3898662052693894) - **Sugon shows off China's first fully domestic 100,000-card AI cluster at WAIC** — Server maker Sugon put its Dawn 8000 super AI cluster on public display at the World AI Conference on Friday, per Pandaily. Every chip, computing unit, storage and networking layer is domestically developed, with a proprietary interconnect that ties all 100,000 accelerator cards into a single fabric, the report said. Sugon has been on the U.S. Commerce Department's Entity List since 2019, blocking its direct access to U.S. chips and design tools. The demonstration anchors Beijing's push to build a domestic training stack that does not rely on Nvidia hardware barred by U.S. export controls. [Pandaily](https://pandaily.com/sugon-dawn-8000-waic-2026-jul2026) # 3. Federal Policy Tracker - **[Congress] Warsh defends Fed AI task force lineup against Sen. Smith's conflict-of-interest challenge** — Federal Reserve Chair Kevin Warsh defended the Fed's AI productivity task force at Wednesday's Senate Banking hearing, calling the three members "the smartest people I know" on AI, per FedScoop. The task force comprises Andreessen Horowitz co-founder Marc Andreessen, Microsoft-owned Xbox executive Asha Sharma and Stanford economist Charles I. Jones, who is on leave at Anthropic. Sen. Tina Smith (D-Minn.) told Warsh that Andreessen has made billions from AI and is a Trump megadonor, Sharma recently oversaw thousands of Xbox layoffs, and Jones is doing paid research for Anthropic. Warsh conceded that Smith's question on whether anyone on the task force has labor market experience was fair. The task force is one of five new independent Fed task forces charged with informing the central bank's policy judgments on AI's economic impact. [FedScoop](https://fedscoop.com/federal-reserve-ai-task-force-kevin-warsh/) [Senate Banking](https://www.banking.senate.gov/hearings/07/08/2026/the-semiannual-monetary-policy-report-to-the-congress) - **[Exec] DOE names Scale AI collaborative hub for Genesis Mission Consortium** — The Department of Energy on Thursday named Scale AI the collaborative hub for Genesis Mission Consortium working groups and structured partnerships, per FedScoop. Scale AI signed a memorandum of understanding with DOE earlier this year, and the new designation expands its role. Emerald AI and SambaNova Systems also joined the consortium in recent weeks, bringing the full private sector roster to Accenture, Amazon Web Services, Databricks, IBM, Emerald AI, SambaNova and now Scale AI. DOE launched the consortium in February 2026 to deepen public-private AI partnerships across the department. [FedScoop](https://fedscoop.com/energy-department-ai-startup-genesis-mission-consortium/) - **[State] New Mexico denies natural gas pipeline for Oracle's Project Jupiter data center for the second time** — New Mexico Commissioner of Public Lands Stephanie Garcia Richard denied Energy Transfer's request to reconsider the state's earlier rejection of two rights-of-way and a business lease for the proposed 24-inch Green Chili Lateral natural gas pipeline, per P&GJ. The pipeline would serve Oracle's Project Jupiter data center in Doña Ana County. In her July 14 letter, Garcia Richard cited high greenhouse gas emissions, water demand and limited benefit to the state land trust, valuing the rights-of-way at nearly $32,000 in one-time payments and the business lease at about $217,000 over five years. The Green Chili Lateral was designed to move up to 400 million cubic feet of natural gas per day to the site. New Mexico now joins Michigan and Palm Beach County in visible state and local pushback against hyperscale AI data centers. Energy Transfer may file a petition for administrative review within 30 days. [Bloomberg](https://www.bloomberg.com/news/articles/2026-07-16/new-mexico-denies-gas-pipeline-permit-for-oracle-data-center) [P&GJ](https://pgjonline.com/news/2026/july/new-mexico-denies-reconsideration-of-project-jupiter-gas-pipeline) - **[State] Hawaii Gov. Green signs deepfake civil remedy law and children's generative AI regulation** — Hawaii Gov. Josh Green (D) this week signed into law bills establishing consumer protection standards for deepfake technology and separately regulating generative AI systems targeted at children, per Maui News. The deepfake law defines harmful use, allows civil action for injunctive relief and punitive damages, and caps monetary damages at $25,000 per piece of content. It covers identity theft, fraud, election interference, cyberbullying and nonconsensual pornography use cases. The child-facing measure applies to conversational generative AI systems used by minors. [Maui News](https://www.mauinews.com/news/local-news/2026/07/gov-green-signs-bills-targeting-deepfakes/) [Hawaii Governor](https://governor.hawaii.gov/newsroom/office-of-the-governor-news-release-gov-green-signs-legislation-to-support-kupuna-care-and-strengthen-ai-protection/) # 4. Capability & Research Watch - **Moonshot's Kimi K3 matches or beats Anthropic's Opus 4.8 on benchmarks** — Moonshot unveiled Kimi K3 on July 16, and its self-reported benchmarks show it matching or surpassing Anthropic's Opus 4.8 on Arena.ai and other tests while trailing OpenAI's GPT 5.6 Sol and Anthropic's Fable 5 in some areas, per TechCrunch and the Financial Times. At 2.8 trillion parameters, it is China's largest open-weight AI model; the Alibaba-backed lab plans to release its weights on July 27. Moonshot is raising a new funding round at a $31.5 billion valuation, up from $20 billion in May. [TechCrunch](https://techcrunch.com/2026/07/16/moonshots-upcoming-kimi-3-is-expected-to-close-the-gap-with-anthropics-opus-4-8/) [SiliconANGLE](https://siliconangle.com/2026/07/16/chinas-moonshot-throws-gauntlet-kimi-k3-worlds-largest-open-weights-model/) - **UN linked Tech Against Terrorism benchmarks 27 AI models and asks governments to geoblock noncompliant systems** — UN supported nonprofit Tech Against Terrorism released a benchmark that tested 27 AI models with prompts for terrorism related information and flagged the threat as a major national security concern, per ABC News. ChatGPT refused most requests but researchers found workarounds. The organization is asking governments to geoblock AI systems that breach national safety laws. A Queensland teenager was charged last month over allegedly using an AI platform to draft mass shooting scenarios, including a note titled "The Albert Massacre" referencing Albert State School in Maryborough. [ABC News](https://www.abc.net.au/news/2026-07-16/terrorist-use-of-ai-major-national-security-concern/106918918) [Tech Against Terrorism](https://techagainstterrorism.org/news/press-release-ai-terrorism-benchmark) # 5. Industry & Market Watch - **4,500 Google workers deliver AWU petition to Pichai for layoff protections** — More than 4,500 Google workers led by the Alphabet Workers Union on Thursday delivered a petition to CEO Sundar Pichai calling for guaranteed severance, mandatory buyouts before layoffs across all product areas and the option to take severance as a buyout, per The Guardian. AWU President Parul Koul said Alphabet's $4 trillion valuation has quadrupled over six years while workers still lack layoff protections. The petition is nonbinding and does not force any action from Google management. [The Guardian](https://www.theguardian.com/technology/2026/jul/16/google-workers-layoff-protections-ai) - **Energy company IPOs raise $12.6 billion in first half of 2026 as investors chase AI power exposure** — Energy company IPOs raised $12.6 billion in the first half of 2026, the highest half year level since the dotcom peak in late 1999 and well above 2025's $4.3 billion full year total, per Ars Technica and Dealogic. U.S. electricity demand is projected to rise 39% between 2026 and 2035, largely due to data center growth, per an estimate from consulting firm ICF. RBC clean energy analyst Chris Dendrinos said investors initially bought AI linked names such as Nvidia before shifting to energy plays needed to power those chips. A typical AI focused data center consumes about 876,000 megawatt-hours per year, roughly the household electricity use of Salt Lake City. [Ars Technica](https://arstechnica.com/information-technology/2026/07/energy-ipos-surge-as-investors-hunt-for-ways-to-play-ai-boom/) # 6. Global & Geopolitics - **EU orders Google to share Search click and query data with rivals and open Android to competing AI assistants** — The European Commission on Thursday, July 16, issued two binding Digital Markets Act specification measures ordering Google to give rival AI assistants and search engines comparable access to Android and to share Search click and query data with competitors, per SiliconANGLE. The measures apply immediately to Alphabet's Search and Android platforms, which each reach billions of users. The Commission issued the orders as guidance on AI interoperability and Search data sharing that Google must implement to remain in DMA compliance. [SiliconANGLE](https://siliconangle.com/2026/07/16/eu-orders-google-share-search-data-rivals-broaden-android-feature-access/) [European Commission](https://digital-strategy.ec.europa.eu/en/news/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data) - **European Commission accepts X's DSA action plan with Board's audit warning still on file** — The European Commission on Thursday, July 16, accepted X's Digital Services Act action plan on advertising transparency and researcher data access. Under the plan, X will improve its advertising repository, publish more ad information and give eligible researchers free access to public data while removing contractual prohibitions on scraping. An independent external audit will assess the changes and report results to the Commission. The Board for Digital Services had earlier deemed X's audit measures and overall action plan insufficient in a June 15 opinion. [European Commission](https://digital-strategy.ec.europa.eu/en/news/commission-accepts-xs-action-plan-comply-digital-services-act)