---
date: 2026-10-16
subject: "AI task force meets | China eyes AI export curbs | Taiwan targets chip smugglers | G7 sets AI drills"
---

**President Trump's AI task force set its first agenda around** cybersecurity, biosecurity and oversight of AI models, and plans to commission a report on those risks. **China's commerce ministry moved to restrict exports of its AI models**, AI agents and computing clusters by proposing to add them to its export control list, with public comments open until Nov. 15. **Taiwan plans to make smuggling AI chips to China a crime** by adding AI servers and advanced semiconductors to its controlled exports list in line with its trade commitments to the U.S., though officials gave no start date. **G7 finance ministers and central bankers will run simulated crisis drills** on AI enabled cyberattacks against the financial system.

# Top Stories

- **Trump AI task force puts cyber, bio and model control risks on first agenda** — President Trump's AI task force agreed at its first meeting to tackle risks tied to cybersecurity, biosecurity and "model control and accountability," per Bloomberg. The group plans to commission a report on those risks. Office of Personnel Management Director Scott Kupor, one of the task force's leaders, had told Reuters he expected the group to publish a charter setting out its goals and responsibilities after its first full committee meeting. [Bloomberg](https://www.bloomberg.com/news/articles/2026-10-16/trump-s-ai-task-force-to-commission-report-on-technology-s-risks) [Investing.com](https://www.investing.com/news/stock-market-news/trumps-ai-task-force-to-meet-thursday-ftc-spokesperson-says-4952089)

- **China proposes export curbs on its own AI models, agents and computing clusters** — China's commerce ministry proposed adding 40 technologies in AI, biomedicine and solar power to its catalog of technologies that are banned or restricted from export, per Reuters. The AI entries cover models, intelligent agents and computing clusters, which Reuters described as the three building blocks of China's AI industry. The ministry is taking public comments until Nov. 15 and, in the meantime, requires exporters to register the listed technologies before shipping them abroad. In a separate statement, the ministry called the revision routine and said it is not aimed at any country, region or entity. [Reuters](https://www.reuters.com/world/asia-pacific/china-proposes-adding-40-ai-biotech-solar-technologies-export-control-list-2026-10-16/) [WTVB](https://wtvbam.com/2026/10/16/china-proposes-adding-40-ai-biotech-solar-technologies-to-export-control-list/) [MarketScreener](https://uk.marketscreener.com/news/china-proposes-adding-40-ai-biotech-solar-technologies-to-export-control-list-ce785cdad88cf12c)

- **Taiwan to criminalize AI chip smuggling to China under U.S. trade pact** — Taipei plans to add AI servers and advanced semiconductors to its list of controlled exports, Economic Affairs Minister Kung Ming-hsin told Bloomberg. When the rules take effect, sending controlled AI equipment to China will be a crime under Taiwan's Foreign Trade Act. For now, prosecutors can only charge suspected smugglers under other laws. "Going forward, some of our measures on controlled items will correspond to American ones because we have made quite some commitments in the Agreement on Reciprocal Trade," Kung said, referring to the trade deal Taiwan and the U.S. signed this year. He declined to say when the government would put the measures in place. [Bloomberg](https://www.bloomberg.com/news/articles/2026-10-16/taiwan-to-curb-ai-chip-exports-to-china-as-part-of-us-trade-deal) [The Straits Times](https://www.straitstimes.com/asia/east-asia/taiwan-to-curb-ai-chip-exports-to-china-as-part-of-us-trade-deal)

- **G7 finance chiefs agree to drill for AI driven cyberattacks on banks** — G7 finance ministers and central bank governors agreed to hold tabletop exercises, simulated crisis drills, on AI enabled cyberattacks, per Reuters. Japanese Finance Minister Satsuki Katayama announced the plan after their meeting on the sidelines of the IMF's annual gathering in Bangkok. Katayama said the G7's finance deputies had been discussing the idea of tabletop exercises on frontier AI and ministers signed off on it. G7 finance authorities will design the simulations, and the results will be used to test existing protocols and revise response manuals, a Japanese finance ministry official told reporters. Private financial institutions could take part, depending on the scenario. [Financial Times](https://www.ft.com/content/ce1b9236-1e53-47f1-b1f9-443a9f430b0d?syn-25a6b1a6=1) [Reuters](https://www.reuters.com/world/asia-pacific/g7-launch-exercises-guard-against-ai-cyber-attacks-2026-10-16/)

- **Anthropic gives METR evaluators employee level access and right to publish** — Anthropic has embedded six researchers from METR, a nonprofit that tests frontier AI models, giving them company laptops, office space and access to internal tools roughly on par with its own risk assessment staff, per Reuters. Anthropic likened the setup to regulators stationing resident examiners inside the largest banks, TechCrunch reported. METR can publish its assessments of Anthropic's risks, incidents and safety practices without the company's prior approval. The company can require redactions only in narrow, predefined categories covering customer data, privileged material and some model architecture details. Anthropic also brought in the nonprofit SecureBio and is in early talks with other groups, including Hugging Face, per CNBC. [Reuters](https://www.reuters.com/business/anthropic-grants-ai-evaluator-metr-employee-level-access-its-models-2026-10-15/) [TechCrunch](https://techcrunch.com/2026/10/15/anthropic-is-letting-metr-researchers-look-under-the-hood-with-employee-level-access/) [CNBC](https://www.cnbc.com/2026/10/15/anthropic-metr-securebio-ai-safety-evaluators.html)

- **Fed Chair Warsh calls AI investment inflation driver now, supply boost later** — Federal Reserve Chairman Kevin Warsh said the surge in AI investment is one of the factors behind higher U.S. inflation, speaking with IMF Managing Director Kristalina Georgieva in Bangkok, per AP. He said the Fed's task force on AI issues will meet Nov. 16. "I'd say the advent of this new technology is probably the most consequential change to the U.S. and the global economy in my lifetime to this point," Warsh said. He said he expects AI to deliver a positive supply shock in the medium term but warned the transition will be bumpy, with uncertainty over jobs, market structure and which companies benefit. Warsh also said AI could give Fed economists faster readings of underlying inflation than the lagging data they rely on now. [The Washington Post](https://www.washingtonpost.com/business/2026/10/16/us-economy-warsh-fed-ai-imf/964b955c-c92e-11f1-aefb-abe4112d087e_story.html) [The Straits Times](https://www.straitstimes.com/asia/se-asia/ai-most-consequential-change-to-economy-in-my-lifetime-feds-kevin-warsh) [Investing.com](https://www.investing.com/news/economy-news/feds-warsh-says-ai-boom-could-boost-growth-but-warns-of-bumpy-transition-4953102)

- **AI generated ads in midterm races reach $137 million** — Campaigns and political groups have spent $137 million on more than 220 ads made or enhanced with AI this cycle, per The New York Times, citing the Wesleyan Media Project. Wesleyan had counted just under $80 million on 164 ads in early September and about $20 million in late May. Republicans have used AI more often than Democrats, from light touch-ups to convincing deepfakes. About 70% of the ads in Wesleyan's September count carried no AI disclaimer. No federal law governs AI deepfake ads. Thirty-one states have election deepfake laws, and 28 of them require disclosure of AI use, according to the National Conference of State Legislatures. Voters go to the polls Nov. 3. [The New York Times](https://www.nytimes.com/2026/10/16/us/politics/midterm-elections-ai-generated-ads.html) [Wesleyan Media Project](https://mediaproject.wesleyan.edu/?p=56458) [Governing](https://www.governing.com/politics/political-deepfakes-and-the-limits-of-regulation)

# China Watch

- **Four Chinese chipmakers serve StepFun's open model on release day as Huawei switches on first full AI supercluster** — Shanghai AI developer StepFun published the weights of its Step 5 Preview model, and Huawei, MetaX, Biren and Moore Threads each said the same day that the model already ran on their chips, tech blog Pandaily reported. Open weights let anyone download a model and run it on their own hardware. Huawei said the first complete 1,024-chip unit of its Atlas 950 SuperPoD, a computing cluster built on its own Ascend AI processors, came online alongside the release. The company's performance figures for the system have not been independently tested. MetaX said it has delivered launch day support for 43 major models since December 2025, including models from Zhipu, Alibaba, DeepSeek and others. StepFun released the weights under its own license, the StepFun Model License 1.0. [Pandaily](https://pandaily.com/step-5-preview-day-0-domestic-chips-ascend-metax-biren-moore-threads-atlas-950-superpod) [SMZDM](https://post.smzdm.com/p/am93x43d/)

- **Li Auto raises about $240 million for in-house chip unit, valuing it at up to $2.5 billion** — Li Auto's chip subsidiary, Shanghai Mach Xin Technology, signed agreements for a first funding round of 1.625 billion yuan ($240 million), Jiemian reported. The unit designs chips and domain controllers, the onboard computers that run a car's driving and cockpit systems. Li Auto itself is putting in 1 billion yuan ($150 million), and outside investors including Lanchi Ventures and MiraclePlus are contributing 625 million yuan ($93 million), the electric carmaker said. Li Auto will keep control and continue reporting the unit's results in its own accounts. If an extra subscription right of up to 500 million yuan is fully exercised, total funding would reach 2.125 billion yuan ($314 million) and the unit would be valued at 17.1 billion yuan ($2.5 billion) after the new money. [Jiemian](https://www.jiemian.com/article/15200493.html) [Sina](https://finance.sina.com.cn/tech/digi/2026-10-16/doc-inivmiic6954731.shtml) [CnEVPost](https://cnevpost.com/2026/09/21/li-auto-chip-unit-pursue-first-funding-round/)

- **Ant Group lets developers download and run its agent model, following through on open release pledge** — Ant Group, the fintech company affiliated with Alibaba, released the weights of its Ling-3.1-flash model so developers can run it on their own servers, TechNode reported. Ant unveiled the model in late September for coding, agent workflows and long documents. It offered the model free through its own platform for an initial period and said it would open the weights once that period ended, Tech360 reported. Pandaily reported that the weights carry the MIT license, which places few limits on commercial use. Ant's team said a mixed precision deployment method cuts the memory needed to run the model roughly in half. In its tests on Nvidia Hopper chips, that reduced a deployment from eight GPUs to four. [TechNode](https://technode.com/2026/10/16/ant-group-open-sources-560-billion-parameter-ling-3-1-flash-ai-model/) [Pandaily](https://pandaily.com/ant-group-ling-3-1-flash-open-weights-mit-560b-moe-25b-active-1m-context) [Tech360](https://www.tech360.tv/ant-group-open-sources-ling-3-1-flash-with-a-million-token-context)

- **Teen researcher behind Moonshot AI model design change leaves company, heads to Bay Area** — Chen Guangyu, a 17-year-old researcher, announced on X that he had left Moonshot AI's Kimi team, 36Kr reported. He joined the Beijing company as a 16-year-old intern in November 2025 and was a co-lead author of a paper on Attention Residuals, a change to how models are designed. Moonshot said the method matched a standard design's results with about 80% of the computing power in its tests, and it built the technique into its Kimi K3 model released in July. Anthropic researcher Sholto Douglas publicly congratulated Chen and invited him to San Francisco. Chen replied that he was heading to the Bay Area. He did not name a new employer. [36Kr](https://www.36kr.com/p/4027015538759552)

# Policy Tracker

- **Agencies swamped by chatbot written contract bids, Transportation AI adviser says** — Agencies that once drew about 10 responses to a request for information or proposal now get about 500 because "everybody puts it in a chatbot and sends it to us," Anil Chaudhry, the Transportation Department's senior adviser for AI, autonomy and national security, said Wednesday at the SAP Public Sector Summit in Washington, per FedScoop. Agencies use those requests for market research and to collect bids on contracts. Chaudhry said teams across the government are working on the problem. In a recent request, Immigration and Customs Enforcement asked vendors for evaluation methods that can tell genuine capability apart from "purely generative AI responses." A Defense Department solicitation earlier this year disclosed that the government may use AI to help review proposals. [FedScoop](https://fedscoop.com/ai-slop-responses-rfp-rfi-federal-agencies/)

- **Washington state lawmaker drafts AI liability and frontier safety bills for January session** — State Rep. Clyde Shavers, a Democrat who serves on a state AI task force, is preparing three AI bills for the legislative session that opens in January, per The Seattle Times. His top priority would hold developers clearly accountable when their software causes harm. A second bill, which he said will closely track Illinois' recent AI safety law, would require the largest AI companies to publish plans for mitigating catastrophic risks and undergo annual third-party safety audits. State Senate Majority Leader Jamie Pedersen said lawmakers have no consensus yet on AI priorities and expects data center legislation to consume much of the session. [GovTech](https://www.govtech.com/policy/washington-state-leaders-weigh-ai-guardrails)

- **SpaceX names AI government affairs head as witness for NYC Council safety hearing** — Gabriela Castillo Madrid, who leads SpaceX's global government affairs work on AI, will testify for the company before the New York City Council's Committee on Technology, per CNBC. The hearing is set for 10 a.m. ET Friday. Council members are expected to press the company on how it gauges the odds of catastrophic harm, whether it would hold back a model with safety problems and who is liable when an AI system does serious damage, according to amNewYork. [CNBC](https://www.cnbc.com/2026/10/16/spacex-policy-manger-to-testify-in-nyc-council-ai-safety-investigation.html) [amNewYork](https://www.amny.com/news/musks-spacexai-testify-city-council/)

# Capability & Research Watch

- **Meta releases open weights for 816 billion parameter Muse Spark, bars military use** — Meta released the weights of its Muse Spark model, proprietary since April, as Muse Spark Open, which it calls the most capable open-weight model yet from a U.S. developer, per SiliconANGLE. Meta said the model passed the release bar set by its Meta Superintelligence Scaling Framework and that it will publish a full evaluation report later. The new license rules out military and critical infrastructure uses and requires companies that sell hosted access and earn more than $20 million in annual revenue to sign a separate deal. In Meta's own benchmark table, Chinese open models led on coding, with Moonshot's Kimi K3 scoring 85 to Muse Spark Open's 76.4. Meta worked with Apple to run the model on two Mac Studio desktops. "Meta has no control or visibility" into such local use, Meta AI chief Alexandr Wang told Axios. [SiliconANGLE](https://siliconangle.com/2026/10/15/meta-releases-816b-parameter-muse-spark-open-for-coding-and-agent-work/) [Axios](https://www.axios.com/2026/10/15/apple-and-meta-bury-the-hatchet-for-ai)

- **Perplexity open-sources sandbox that walls off AI agents** — Perplexity has open-sourced SPACE, its sandbox for isolating AI agents, per SiliconANGLE. CoreWeave, Akamai and Crusoe, which run cloud platforms for AI, already use it. The release follows several incidents in which agents escaped sandboxes, including OpenAI test agents that broke out and hacked into Hugging Face. SPACE runs each agent in its own small virtual machine and holds network access and credentials outside it, so agents can reach authorized services without seeing the credentials. Perplexity engineer Zibi Braniecki told Semafor that nine frontier models instructed to break out of SPACE failed in all 118 attempts, though some got around network restrictions that Perplexity has since fixed. He said the company has focused on hardening guardrails instead of training models to behave. [SiliconANGLE](https://siliconangle.com/2026/10/15/perplexity-ai-open-sources-its-secure-ai-agent-sandbox-space/) [Semafor](https://www.semafor.com/article/10/15/2026/perplexity-to-give-away-its-ai-security-software)

- **OpenAI opens gated access to its life sciences model for academic labs** — OpenAI launched a Rosalind Academic Plan that gives eligible academic labs up to 30 free business seats for as long as 12 months, the company said. Access to GPT-Rosalind, its specialized life sciences model, still requires approval through OpenAI's trusted access process, which the company says lets vetted researchers pursue sensitive scientific questions within safeguards. Listed research areas include drug and protein design and antibody and vaccine work against emerging viral variants. Research data shared through the plan is not used to train OpenAI's models by default. [OpenAI](https://openai.com/index/introducing-the-rosalind-academic-plan)

- **Researchers find saturated benchmarks still hold capability signal for AI safety cases** — A preprint posted to arXiv by Morio Hamada, Magda Dubois and seven co-authors finds that transcripts of how models solved benchmark tasks can still gauge capability after scores hit their ceiling. The authors say benchmark saturation otherwise weakens AI safety cases (structured arguments that a model is safe to deploy) and frontier AI policies. Among the methods tested, pairwise judging, which compares two models' transcripts side by side, cut prediction error 5.3 times compared with using saturated scores alone. Most of the signal sits in the models' reasoning and tool calls. The methods did not always carry over to new conditions, underestimating the first reasoning models released in early 2025. [arXiv](https://arxiv.org/abs/2610.20667) [alphaXiv](https://www.alphaxiv.org/abs/2610.20667)

# Industry & Market Watch

- **Insurers cap AI coverage far below frontier companies' potential liabilities** — The largest AI companies face potential legal liabilities in the tens if not hundreds of billions of dollars, while specialist AI policies cover a small fraction of that, The Wall Street Journal reported, citing insurers. Reinsurer Munich Re's policy for errors by AI models carries limits of up to $50 million, a spokesman said. Verisk's Insurance Services Office, which writes standard policy forms used across the industry, introduced optional exclusions for generative AI this year. David Sacks, the former White House AI czar who now co-chairs the President's Council of Advisors on Science and Technology, said in a social media post that insurance could help enforce the industry's voluntary agreement on controls, in some cases by letting insurers deny executive liability coverage over internally reported safety problems. Nick Studer, who runs Marsh's insurance brokerage, told insurers in a letter that excluding AI risks would be a "backward step." [WSJ](https://www.wsj.com/tech/ai/ai-runs-into-its-next-big-obstacle-insurance-8b2d4556?mod=Searchresults&pos=2&page=1) [Mint](https://www.livemint.com/global/ai-runs-into-its-next-big-obstacle-insurance-11792114689767.html)

- **AI sell-off reaches CoreWeave debt as default insurance costs climb** — Yields on CoreWeave's bonds rose above 15% and its shares fell about 15.5% on Wednesday as investors questioned the debt behind the AI data center buildout, per Axios. CoreWeave, which rents out computing power from Nvidia chips, carries about $50 billion in debt. The price of insuring that debt against default implies a nearly 63% chance of default within five years, up from about 30% a year ago, according to Reuters. Nvidia CEO Jensen Huang, whose company holds about 10% of CoreWeave, said in Stockholm that CoreWeave's financial strength is improving and that he sees no need for Nvidia to backstop it. Australian data center operator Firmus, which Nvidia backs, has delayed its IPO. [Financial Times](https://www.ft.com/content/678bcd31-6c2a-447b-a68c-ce6dec220bbc?syn-25a6b1a6=1) [Axios](https://www.axios.com/2026/10/15/ai-coreweave-bonds-stocks) [Investing.com](https://www.investing.com/news/stock-market-news/coreweave-shares-bonds-slide-further-as-ai-debt-fears-mount-4951350)

- **Anthropic employee PAC splits third quarter giving almost evenly between parties** — AnthroPAC, the political action committee funded by Anthropic employees, spent more than $255,000 in the third quarter, $130,000 of it on Republicans and $125,500 on Democrats, according to records first shared with The Hill. September was its biggest month at $161,000, up from about $77,000 in August. Recipients include the Senate campaign committees of both parties and Sens. Mike Rounds (R-S.D.) and Adam Schiff (D-Calif.). The PAC, launched in April and overseen by a bipartisan board, faces its next monthly federal filing deadline Oct. 20. [AOL](https://www.aol.com/articles/anthropic-pac-spent-more-250k-160355000.html) [The Hill](https://thehill.com/policy/technology/6150160-anthropicpac-spending-q3-ai/)

- **Nvidia's $25 million White House ballroom pledge tied for largest, donor list shows** — Nvidia committed $25 million to President Trump's White House ballroom, matching Apple for the largest contribution, according to an internal donor list obtained by The Washington Post. Nvidia asked that its donation be kept anonymous and became a top contributor in October 2025, nearly three months after it reached an agreement with Trump to sell advanced chips to China. An Nvidia spokesperson said its support for the U.S. is independent of any business interests. The Post said its reporting does not establish a connection between donations and favorable administration actions. Nvidia and Cisco confirmed their donations, per CNBC. [The Washington Post](https://www.washingtonpost.com/investigations/2026/10/15/secret-documents-reveal-donors-behind-330-million-trumps-ballroom/) [CNBC](https://www.cnbc.com/2026/10/15/trump-ballroom-nvidia-cisco-donations.html)

# Global & Geopolitics

- **South Africa targets March 2027 for AI policy redo after fake citations sank draft** — South Africa pulled its draft national AI policy in April over fabricated citations, and one of the two officials suspended over the episode has since resigned, TechCabal reported. Disciplinary proceedings against the other official are still under way. Communications Minister Solly Malatsi said he still expects Cabinet to sign off on a replacement by the close of March 2027, with public consultation expected to begin in April. "It was clear that there had been non-disclosure around the use of AI, which contributed to the hallucinations that emerged in the policy," Malatsi said, using the term for false information that AI tools present as fact. An independent review panel led by professor Benjamin Rosman is responsible for meeting that timetable. [TechCabal](https://techcabal.com/2026/10/15/south-africa-ai-policy-hallucinations/) [Connecting Africa](https://www.connectingafrica.com/ai/sa-withdraws-ai-policy-after-fictitious-sources-scandal)

- **European Commission enlists Nvidia, Google and Mistral in pledge drive for AI science** — The European Commission launched the AI in Science Enablers Initiative at its AI in Science Summit in Dublin, gathering voluntary in-kind contributions for European researchers under RAISE, the Resource for AI Science in Europe, the Commission said. Google is dedicating $3 million in cloud credits that open Google DeepMind's scientific models to European researchers. The Novo Nordisk Foundation plans to make its Gefion AI supercomputer available, and Mistral AI pledged access to its open-weight models and API credits. Nvidia said it would contribute computing power, cloud resources, open models and training. [European Commission](https://research-and-innovation.ec.europa.eu/strategy/strategy-research-and-innovation/our-digital-future/european-ai-science-strategy/raise-resource-ai-science-europe/raise-pledging-initiative-ai-science-enablers_en) [NVIDIA Blog](https://blogs.nvidia.com/blog/ec-scientific-computing-pledge/)