Taiwan is preparing tighter export controls on AI chips and servers destined for China, with Economic Affairs Minister Kung Ming-hsin confirming the changes follow commitments under the 2026 Taiwan-US Agreement on Reciprocal Trade. The government plans to add AI hardware to a list of controlled items and make unauthorized shipments a criminal offense under the Foreign Trade Act. The move targets ongoing smuggling of restricted Nvidia and AMD chips into China, including a case where Super Micro Computer employees faced a DOJ investigation for allegedly smuggling $2.5 billion in Nvidia chips.
In the corporate AI space, Anthropic disclosed that rival labs' distillation of its Claude outputs accounted for 5 to 7 percent of its revenue. The company tightened access to Claude in the third quarter, reducing what competitors could harvest by roughly 90 to 99 percent. Anthropic named Zhipu's GLM-5.3 and Moonshot's Kimi as beneficiaries of previously extracted data.
Investor letters shed light on how major firms are positioning around AI. Wedgewood Partners praised Microsoft for growing operating income by about 18 percent and maintaining returns on invested capital near 30 percent, generating $35 billion in incremental gross cash flow on $115 billion in average incremental assets. Emerald Wealth Partners highlighted ServiceNow, noting its systems are deeply embedded in customer businesses, making replacement too costly and risky for large organizations. ServiceNow shares rose 35.0 percent in the covered period.
On the technical front, StepSafe introduced ASPO, a framework for step-level safety reasoning in agentic AI systems, achieving a Youden's J of 53.0 at 4B parameters on agent-safety benchmarks. Separately, the Postgres core team delayed Postgres 19 to November after AI reviewers found more bugs than human developers could fix, reverting up to 10 features during the process. Apple also updated its App Store rules to explicitly ban nudify AI apps that create nonconsensual intimate imagery, along with apps facilitating human trafficking or child exploitation.
Key Takeaways
- Taiwan plans to criminalize unauthorized AI chip shipments to China under the Foreign Trade Act, aligning with US restrictions on Nvidia and AMD hardware.
- Anthropic said rival distillation made up 5-7% of its revenue before it tightened Claude access, cutting what competitors could harvest by 90-99%.
- Microsoft grew operating income by about 18% and generated $35 billion in incremental gross cash flow, per Wedgewood Partners' Q3 2026 letter.
- ServiceNow shares rose 35.0% as investors cited its deeply embedded enterprise systems that are too costly to replace.
- StepSafe's ASPO framework achieved state-of-the-art safety performance with a Youden's J of 53.0 at 4B parameters for agentic AI systems.
- Postgres 19 was delayed to November because AI reviewers found more bugs than human developers could fix.
- Apple updated App Store rules to ban nudify AI apps, apps facilitating human trafficking, and child exploitation content.
- Super Micro Computer employees faced a DOJ investigation for allegedly smuggling $2.5 billion in Nvidia chips to China.
- Anthropic named Zhipu's GLM-5.3 and Moonshot's Kimi as beneficiaries of data extracted through distillation of Claude outputs.
- Smuggling of restricted AI hardware via Southeast Asia continues to be a concern, with Nvidia chips still reaching China despite US restrictions from 2022.
Taiwan to tighten AI chip exports to China under US trade deal
Taiwan plans to add AI servers and advanced semiconductors to a list of controlled items. Economic Affairs Minister Kung Ming-hsin said the move aligns with the 2026 US trade agreement. Shipping restricted AI hardware to China will become a criminal offense under the Foreign Trade Act. The government has not set a timeline for the new rules.
Taiwan mulls stricter AI chip export controls to align with US
Taiwan is considering tighter export controls on AI chips sold to China. The move targets smuggling risks and would give authorities more legal tools. The focus is on AI servers using Nvidia chips diverted from Taiwan to China. The plans were reported by Bloomberg citing people familiar with the matter.
Taiwan weighs stricter AI chip sales curbs to China: Bloomberg
Taiwan is reviewing stronger export controls on AI chip sales to China as part of closer alignment with the US. The measures aim to address chip smuggling and give authorities legal tools to stop advanced hardware from reaching China. Nvidia chips have still reached China through smuggling via Southeast Asia despite US restrictions from 2022. Super Micro Computer employees faced a DOJ investigation for allegedly smuggling $2.5 billion in Nvidia chips.
Taiwan plans tighter controls on AI chip exports to China
Taiwan plans to tighten controls on advanced AI chips and servers exported to China. Economics Minister Kung Ming-hsin said the changes follow commitments under the Taiwan-US Agreement on Reciprocal Trade. Some export controls would mirror US rules, though no implementation date was given. The move comes as concerns grow over smuggling of restricted AI equipment into China.
Taiwan to criminalize unauthorized AI chip shipments to China
Taiwan plans to make unauthorized shipments of restricted AI chips and servers to China a criminal offense. The change would allow authorities to prosecute smugglers directly under the Foreign Trade Act. Currently, prosecutors must rely on other existing laws to pursue such cases. The move aligns with US restrictions on Nvidia and AMD chips and follows the 2026 Agreement on Reciprocal Trade.
ServiceNow's deep system roots shield it from AI disruption
Emerald Wealth Partners highlighted ServiceNow (NYSE:NOW) in its Q3 2026 investor letter. The firm said ServiceNow's systems are deeply embedded in customer businesses with data, permissions, and controls. Large organizations find it too costly and risky to replace such infrastructure. ServiceNow shares rose 35.0% in the period covered by the letter.
Microsoft remains a high-quality AI leader, investors say
Wedgewood Partners praised Microsoft (NASDAQ:MSFT) in its Q3 2026 investor letter. The firm noted Microsoft grew operating income by about 18% and maintained returns on invested capital near 30%. Microsoft generated $35 billion in incremental gross cash flow on $115 billion in average incremental assets over 12 months. The firm expects more upside as Microsoft reaps returns on its AI capital spending.
Protecting enterprise data is key to securing AI's future
Proofpoint argues that protecting enterprise data is essential as AI reshapes the workplace. The rise of AI-powered tools expands the attack surface for data breaches and cyber attacks. Organizations need data encryption, access controls, and real-time monitoring. AI-powered security solutions can help detect and respond to threats quickly.
StepSafe introduces step-level safety reasoning for AI agents
StepSafe is a new framework called ASPO for step-level timely intervention in agentic AI systems. It evaluates each decision at its own step and rewards actions taken before risks become visible. The framework uses adaptive feedback self-distillation with the model as its own teacher. StepSafe achieved state-of-the-art performance with a Youden's J of 53.0 at 4B parameters on agent-safety benchmarks.
Anthropic says rival labs' distillation made up 5-7% of revenue
Anthropic told investors that distillation by rival AI labs accounted for 5 to 7 percent of its revenue. The company cut this back materially in the third quarter by tightening access to Claude. Distillation occurs when competitors train models on Anthropic's outputs. Anthropic's defenses reduced what distillers could harvest by roughly 90-99%. The company named Zhipu's GLM-5.3 and Moonshot's Kimi as beneficiaries of extracted data.
Seven AI stocks to watch according to MarketBeat
MarketBeat listed seven artificial intelligence stocks worth watching. They include Tempus AI, Hut 8, SentinelOne, C3.ai, Upstart, Appian, and SoundHound AI. These are companies that develop or use AI technology.
Postgres 19 delayed as AI finds more bugs than humans
The Postgres core team delayed Postgres 19 to November because AI reviewers found more bugs than human developers could fix. The team reverted up to 10 features during the process. The delay was caused by better AI tooling, not by vibe coding practices.
Apple updates App Store rules to ban nudify AI apps
Apple updated its App Store rules to explicitly ban apps that create nonconsensual intimate imagery. The new rules also target apps that facilitate human trafficking or child exploitation. Such apps will be pulled from the store and blocked from alternative distribution.
LEV decision model improves accuracy through looping
Researchers introduced LEV, a decision model that thinks deeper by rerunning its layers in a loop. A 2.6B LEV improved from 63.2% accuracy at depth one to 76.2% at depth four on JevBench. On-demand depth settings save 15% to 24% of computation with minimal accuracy loss.
OTA listings now shape how AI recommends hotels
Online travel agencies now drive over 70% of travel bookings and shape AI hotel recommendations. A study found 75% of hotel listings on OTA platforms contained errors. Hotels with accurate listings saw a 25% increase in bookings compared to those with inaccurate ones.
Sources
- Taiwan to curb AI chip exports to China as part of US trade deal
- Taiwan to curb AI chip sales to China for more U.S. alignment- Bloomberg
- Taiwan to curb AI chip sales to China for more U.S. alignment- Bloomberg
- Taiwan to impose strict controls on AI chip exports to China | Taiwan News
- Taiwan to make shipping AI hardware to China a criminal offence
- How Deep System Roots Protect ServiceNow (NOW) in AI Era
- Microsoft (MSFT) Remains a High-Quality AI Leader
- Securing the AI Future Starts with Protecting Enterprise Data
- StepSafe: Learning Timely Intervention with Context-Aware Safety Reasoning
- 5-7% Of Our Revenue Was Because Of Rival Labs’ Distillation, Anthropic Tells Investors
- Top Artificial Intelligence Stocks Worth Watching
- Postgres 19 delayed after AI reviewers find more bugs than humans can fix
- Apple’s updated App Store rules go after ‘nudify’ AI apps
- LEV: Thinking Fast and Deep with Looped Decision Models
- The Global Consultant: OTA listings now shape how AI recommends hotels
Comments
Please log in to post a comment.