Latest / HotTea: The AI Morning Briefing / Four AI firms agree to testify as New York subpoenas SpaceXAI
Transcript
- Sections 00:00 What we're covering today 00:35 1. Four AI firms agree to testify as New York subpoenas SpaceXAI 01:24 2. Nvidia puts AI agent permissions outside the agent 02:19 3. VSMC opens Singapore chip plant, targets volume production in 2027 03:10 4. NaiveAI releases a 309-billion-parameter model with open weights 04:02 5. Washington and Beijing plan a channel for AI incidents 04:49 Visit Hot Tea Disclosure Narration uses an AI-generated voice. Transcript Welcome to Hot Tea for September twenty-eighth. New York City is pressing A I firms to answer questions under oath. We'll start with the planned hearing, then turn to new safeguards for software agents and a chip plant that has opened in Singapore. We'll also cover an open model with a steep memory requirement and plans for a U S-China incident channel. Some of these systems are working now. Others are still company targets or diplomatic commitments. This narration uses an A I-generated voice. Four A I firms agree to testify as New York subpoenas SpaceXAI. New York City Council Speaker Julie Menin said on September 28 that four A I firms had agreed to testify under oath. Anthropic, OpenAI, Google and Meta are due at an October 5 hearing. The council says SpaceXAI didn't respond, so it issued a subpoena. If the company doesn't comply, the council may ask a state court to enforce it. For now, the testimony is only planned, and it concerns proposed laws. The hearing hasn't happened, and the bills haven't become law. October 5 will show which company representatives appear and what they commit to under oath. Any later changes to the bills will show whether their testimony made a difference. the chipmaker puts A I agent permissions outside the agent. Nvidia detailed OpenShell 0.1.0 on September 28. The open-source software limits which files and services an agent can use. A separate supervisor checks each request against rules set by the operator. Nvidia says an operator can allow an agent to query a service while blocking write requests to it. The public release history already lists version 0.1.1, which makes a network tunnel opt-in. The cited sources explain the controls, but they do not include an independent security audit. The policy checker can enforce the permissions an operator sets. That does not prove the rest of the system stays secure while the agent runs. Independent testers need to try blocked writes, child processes and network access. The restrictions need to hold when an agent tries another route. VSMC opens Singapore chip plant, targets volume production in 2027. NXP and Vanguard International Semiconductor opened their VSMC plant in Singapore on September 28. NXP says construction is complete and trial production has begun. The plant uses 130-to-40-nanometer processes to make specialty components, including power-management chips and interposers used in packaging. NXP expects volume production in the first quarter of 2027 and about 44,000 wafers a month by 2029. Those are company targets. The plant is open, but it hasn't reached full capacity. It will make specialty components rather than the newest leading-edge processors. Customers still need to qualify the plant before volume production can begin in early 2027 as planned. Sustained output will matter more than the opening ceremony. NaiveAI releases a 309-billion-parameter model with open weights. NaiveAI released Naive-N0.5-Flash with model weights and inference code under the MIT license. Its model card lists 309 billion total parameters, with 15.5 billion active for each token. The company says the model supports a million-token context through local and sparse attention. It builds on Xiaomi's MiMo-V2.5 base model. Open weights don't make this a small desktop model. The company says the compressed weights alone take roughly 315 gigabytes, before the extra memory needed to run it. The company is still the only source for the performance figures. Independent coding tests need to measure performance and memory use with long contexts. Buyers also need to know the cost of each completed task, including retries and review. Washington and Beijing plan a channel for A I incidents. A September 25 White House fact sheet says the United States and China established an A I dialogue. The next exchange is due by November 2026. UPI reported the agreement on September 26. The two countries also agreed to create an incident channel for what the statement calls super intelligence. That's a diplomatic commitment, not proof that the channel is operating. The statement doesn't set public reporting deadlines or explain which incidents must be shared. The two countries still need to name contacts and define which incidents must be reported. They also need to publish a date for the next meeting. Until then, the channel remains a diplomatic commitment. That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.