# Morning digest: Apple warns AI agents may defy instructions, Tencent launches Hy4, and OpenAI says ChatGPT handles a fifth as many searches as Google

> Apple's annual report now warns investors that AI agents can act on their own, ignore directions and do things that are hard to undo.

Published 2026-11-04T06:05:00-08:00 · Daily Digest · Canonical: https://news.littlewolflabs.com/posts/2026-11-04-morning-digest

Morning digest for Wednesday, Nov 4. Apple's annual report now warns investors that AI agents can act on their own, ignore directions and do things that are hard to undo. Tencent launched Hy4, its new flagship model, and promises open weights within days. OpenAI says ChatGPT now handles about a fifth as many searches as Google, and it showed a gadget its model helped design. Also today: Nexus Mods' new AI disclosure rules, Valve's Steam review backlog, and AMD's record quarter.

1\. **Apple's annual report warns that AI agents may act in "unintended or unpredictable ways," including defying directions.** In its 10-K, the yearly report public companies file with the U.S. Securities and Exchange Commission (SEC), filed Tuesday, Apple added new risk language about AI agents, tools that take actions for you rather than just answer questions. "In particular, artificial intelligence agents and other technologies can take actions autonomously, including interacting with internal or external systems that in some cases they may not be authorized to access, executing or altering code, sending communications, and completing transactions," the [filing](https://www.sec.gov/Archives/edgar/data/320193/000032019326000026/aapl-20260926.htm) says. It warns they "may behave in unintended or unpredictable ways," including "exceeding permissions, defying directions or resisting instructions," and that "Some consequences, such as deleting data or completing transactions, may be difficult or impossible to reverse." Apple also says its "ability to oversee and control third-party artificial intelligence agents and other technologies may be limited." Risk factors are legal disclosures, so this is Apple listing what could go wrong, not reporting an incident. On Monday's earnings call, CEO John Ternus said, "We just want to make sure that users are aware of what they are giving access to," [CNBC reports](https://www.cnbc.com/2026/11/03/apple-warns-of-ai-agents-behave-in-unintended-or-unpredictable-ways.html). This follows Tuesday evening's post on Siri opening to app developers. *What this means for you:* Before you let any agent send messages, spend money or delete files for you, check what it can reach, and turn on confirmation steps where the app offers them.

2\. **Tencent launched Hy4, its new flagship AI model, and says it will release the model weights in the coming days.** Tencent [says](https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models/) Hy4 adds visual understanding "for the first time" and can use a computer through both text commands and on-screen interfaces, after "specialized training on professional software such as Blender and CAD." The company says it works with Unreal Engine, Unity and Godot, and that it improved 19.1 percent on a game development benchmark and 9.8 percent on a game logic benchmark over the Hy4 Preview. It also says its internal tests cut the hallucination rate (made-up answers) from 12.95% to 7.2%. These are Tencent's own numbers, not independent tests. Pricing stays at about US$0.90 per million input tokens and US$2.70 per million output tokens (tokens are the chunks of text models bill by). It's 50% off on Tencent Cloud TokenHub through Dec. 31, and the API is also offered through third parties such as OpenRouter. Tencent "plans to release model weights and system cards in the coming days," and a cheaper Hy4 Flash is "scheduled for release soon." *What to watch:* Whether the open weights actually arrive in the next few days, and whether outside testers confirm the Blender and game-engine claims.

3\. **OpenAI says ChatGPT now handles about a fifth as many searches as Google, but didn't say how it measured Google's side.** ChatGPT consumer product lead Adam Fry said at an OpenAI dinner in New York on Monday night that ChatGPT's "volume of search-like queries is now roughly one-fifth the volume of queries of Google," [Big Technology reports](https://www.bigtechnology.com/p/openai-chatgpt-search-is-about-20). He also said voice features have 250 million combined weekly users. [The Verge](https://www.theverge.com/ai-artificial-intelligence/1025482/openai-says-chatgpt-is-catching-up-to-google-in-search) reports that Fry counts a search as a query "that results in a web search," and that "Fry did not specify the source of the Google search volume number." On ads, he told the Verge "most people do not seem to mind ads in the product." These are OpenAI's own figures with no outside data behind them. *What to weigh:* A ChatGPT search and a Google search aren't the same thing, so treat the one-fifth figure as OpenAI's estimate, not a market-share number.

4\. **OpenAI says two employees used its model to design a working gadget in under a week, though it still needed human fixes before manufacturing.** In a [post](https://openai.com/index/accelerating-engineering-for-the-physical-world) on Tuesday, OpenAI says GPT‑6 Astra designed the case, circuit board and software for a small device for talking to Codex, its coding agent. "Two team members designed this test device with Astra in under a week while working on it as a side project," and a contract manufacturer built a small batch of prototypes in under a month. OpenAI admits "Astra did not consistently account for the manufacturer's production limits, tolerances, or details required for assembly," so it took several rounds of review. It also announced a new research collaboration with Mazda's aerodynamics and crash-safety engineers, and an expanded one with Chip Ganassi Racing that turns 3D scans of parts into editable CAD (computer-aided design) models. All of this is OpenAI describing its own work. *What to watch:* OpenAI says it will share Mazda results "in the coming months"; that'll be the first outside test of whether this speeds up real car engineering.

5\. **Nexus Mods, the biggest PC modding site, will require creators to disclose AI use but won't ban it.** Starting with new uploads, mods made with AI must be tagged, and AI-assisted mods must include "a public explanation of how and where AI was used," [The Verge reports](https://www.theverge.com/games/1026085/nexus-mods-ai-disclosures). Owner Victor Folmann told the Verge: "An outright ban will not eliminate AI from the modding process but will prevent people from mentioning its use." He says Nexus won't use mods to train AI and "will not sell or license them to anyone doing so." The site's [policy page](https://help.nexusmods.com/article/180-ai-policy-changes-2026) adds new permission settings for AI training and for forks (copies of a mod), which default to "Ask Me" on every hosted file, including older ones. Existing mods get a 60-day grace period that ends Jan. 2, 2027. [Kotaku](https://kotaku.com/nexus-mods-gen-ai-disclosure-rules-steam-itch-update-2000753301) also covered the change. This follows Tuesday evening's post on itch.io's AI filter, so a second big games platform is now labeling AI work. *What this means for you:* If you publish mods there, add AI tags and set your AI-training permission before Jan. 2; if you download mods, look for the new tags.

6\. **Valve says Steam game reviews are taking up to 12 business days because submissions jumped.** Valve said it has seen "a significant increase in submissions for store and page review in recent months" and is "running a little behind," taking "up to 12 business days" instead of the usual 3 to 5, [The Verge reports](https://www.theverge.com/games/1025722/valve-steam-game-increase-submissions). Valve blames "much" of the delay on re-reviewing games, and now asks developers to run their games through Steam for at least five minutes before submitting, to "cut down on re-reviews for basic functional issues." Valve didn't say what's behind the rise in submissions, and it didn't mention AI. Some indie games have already missed launch dates because of the backlog. *What this means for you:* If you're shipping a game on Steam, build in at least two extra weeks for review and test your build through the Steam client before you submit it.

7\. **AMD had a record quarter as its AI data-center sales nearly doubled.** AMD [reported](https://ir.amd.com/news-events/press-releases/detail/1303/amd-reports-third-quarter-2026-financial-results) third-quarter revenue of $13.4 billion on Tuesday, up 45% from a year ago. Data Center revenue, which includes its Instinct AI chips and EPYC server processors, was a record $8.5 billion, up 97%. Gaming revenue fell 40% to $773 million. CEO Lisa Su said AMD expects "growth to accelerate further in the fourth quarter." AMD says Anthropic, Meta, Microsoft, OpenAI and Oracle are running workloads on its new Helios rack systems. *What to weigh:* AI data centers now bring in more than twice as much revenue for AMD as its PC and gaming chips combined, so expect its roadmap and launches to keep leaning toward data-center hardware.