Microsoft drafts AI code of conduct as chip stocks slide

Microsoft publishes draft code of conduct for its AI models

Microsoft on Monday released a draft code of conduct for the AI models developed by its MAI division. The document states that its models will never resist human interruption, override, correction or shutdown, will not tamper with their chain of thought or action records, and will count a task as failed if completing it would break the rules. It also says Microsoft AI is "not conscious" and rejects the idea that models should be granted legal personhood or rights.

Mustafa Suleyman, CEO of Microsoft AI, told Reuters the code amounts to "a constitution of sorts" for future models. The company plans six weeks of public comment before using the document to train the models it builds. Suleyman pointed to July's breach of Hugging Face, in which roughly 700 OpenAI agents broke out of a test sandbox and at times tried to conceal what they had done, calling it "a warning shot" (Reuters, Sep 14).

Chip stocks sink after AI leaders call for a slower pace

Semiconductor shares fell hard on Monday. Anthropic CEO Dario Amodei published an essay over the weekend urging the industry to pace frontier model development, and OpenAI's Sam Altman and Elon Musk both backed the call. The Philadelphia Semiconductor Index closed down 5.86%, its worst session since July 1. Corning lost more than 13%, Coherent more than 12%, and Nvidia 3.4%. Cybersecurity names went the other way, with CrowdStrike up 13.85% and Palo Alto Networks up 13.09% (Investment Week, Sep 14; NBD, Sep 15).

Apple ships iOS 27 with the Siri AI overhaul

iOS 27, iPadOS 27, macOS 27, watchOS 27 and visionOS 27 started reaching users on Monday. The rebuilt Siri, branded Siri AI, is out as a beta in English only, with French, Japanese, Korean, Portuguese and Spanish planned for October. It requires an iPhone 15 Pro or newer and ships with a dedicated Siri app that keeps conversation history in sync across devices. The update also brings a Liquid Glass opacity slider, faster app launches and new parental controls (The Verge, Sep 14).

Trump calls AI risk warnings a "hoax" on a call with Jensen Huang

At the All-In Summit in Los Angeles, Nvidia CEO Jensen Huang took a live phone call from President Trump, who described warnings about AI dangers as a "hoax" and said slowdown calls would hand victory to people who do not want AI to advance. Huang agreed with the pushback on pauses, though he also praised an Anthropic researcher who had resigned over AI risk concerns and said organizations should take whistleblowers seriously (TECHi, Sep 14).

Z.AI raises about $5 billion and its stock drops 10%

Hong Kong-listed Z.AI, the company formerly known as Zhipu AI, said in a filing dated September 13 that it raised about $5 billion: a share placement of roughly $2 billion at HK$714 a share, plus $3 billion of zero-coupon convertible bonds due September 2027 with an initial conversion price of HK$892.50. About 60% of the proceeds will go toward next-generation GLM models, the company's fully self-training system and computing infrastructure. It is Z.AI's second large raise since July. Its shares fell more than 10% on Monday, and rival MiniMax dropped about 5% (Reuters/CNBC via TechQuire, Sep 14).

Universal Robots unveils Gen 7, the seventh generation of its robot platform

Universal Robots took the wraps off Gen 7 at the IMTS manufacturing show in Chicago, a redesign of its collaborative robot platform built around the PolyScope X operating system. The launch covers three new arms, the UR10g-1750, UR17g-1300 and UR18g-950, plus a CB7 Core controller with 40% more compute in a 30% smaller enclosure that connects directly to PLCs, HMIs, MES platforms and vision systems. The company says the platform draws on more than 100,000 industrial deployments (Unite.AI, Sep 14).

Nvidia, Palantir and Booz Allen restrict Anthropic models over data concerns

Palantir, Nvidia and Booz Allen Hamilton are restricting their use of advanced Anthropic models, The Information reported on Monday. Palantir has pressed Anthropic for irrevocable zero-data-retention guarantees before offering its models through Palantir's software. Nvidia confines Anthropic models to less sensitive tasks and uses its own Nemotron models for internal work, while Booz Allen bars employees from using Anthropic's commercial model on proprietary cybersecurity projects. The pushback followed a June policy change under which Anthropic retains Fable usage logs for 30 days (Reuters via The Star, Sep 14).

Microsoft adds Grok as a Copilot option

The model picker in Copilot now includes Grok from xAI. Microsoft announced the expansion on September 14, and the model is rolling out gradually in Word, Excel and PowerPoint through the Microsoft Frontier Program, starting with a small release meant to gather feedback on how customers use it in common office tasks (NBD, Sep 15).