AI News This Week: 9 to 13 November 2026, With So-Whats

The AI story that matters most to marketers in the week of 9 to 13 November 2026 is OpenAI's beta of verified business profiles and branded agents inside ChatGPT, announced on 10 November. The rest of the week brought new Google inference chips, a chipmaker buying into AI labs, a metered AI tier from Apple, Australia's regulator plan and a multi-agent research warning. Every item below links to its source.

Key Takeaways

• OpenAI's "Businesses in ChatGPT" beta lets verified businesses claim a profile, see how often ChatGPT mentions them, and run agents. Agents are limited to US e-commerce and online-service firms for now.
• Google announced TPU 8tx and TPU 8i Max on 9 November. A trade-press analysis says both are slated for 2027, so they do not change your costs this year.
• TSMC's board approved up to $500 million each in Anthropic and OpenAI on 10 November, according to the company's own filing.
• Apple launched a $49.99 Intelligence Usage Pack on a limited basis, a sign that metered AI is reaching mainstream software.
• Australia's government said on 10 November that it will legislate AI safety rules and create a regulator. Nothing is law yet.
• Anthropic researchers published simulated-environment results showing groups of agents, one per user, performed worse than a single coordinating agent.
• I left out model launches and Nvidia's leadership change, which I cover separately, and I flag below one item I could not verify.

The week at a glance

1. OpenAI adds business profiles and agents to ChatGPT (10 November)

The Next Web reported that OpenAI announced "Businesses in ChatGPT" in beta on Tuesday. Verified businesses get a blue check and a tool called Business Hub, where they can update details, prices and stock. Business Hub also shows how often ChatGPT mentions the business, how many users click through to its website and common conversation topics. OpenAI says it does not share users' chats with businesses.

Eligible US e-commerce and online-service companies can also launch a Business Agent. When the business appears in an answer, a user can open its profile and chat with the agent. The report says agents must disclose they are AI, keep prices accurate and avoid pressure tactics, and purchases need explicit user approval. Checkout still sends buyers to the brand's site, and OpenAI is working on agent payments with Visa. The same agent can be sold as an ad; the report cites Digiday saying OpenAI states that paying does not change how often ChatGPT recommends a business. The 1.2 billion weekly user figure is OpenAI's claim.

So what: this is the first first-party visibility data from ChatGPT, so claim your profile as soon as you are eligible and treat the mention counts as the baseline for your AI share-of-voice reporting. See my post on ChatGPT's 1.2 billion weekly users.

2. Google unveils TPU 8tx and TPU 8i Max (9 November)

The Next Platform describes TPU 8tx as a variant aimed at inference, with smaller pods of 1,152 accelerators instead of 9,600, a different interconnect and AMD host processors. TPU 8i Max puts two chips on each socket. The author says both are slated for 2027, and the pricing estimates in the piece are the author's speculation, not Google figures. Another outlet in my search swapped the chips' roles, so I trust the Next Platform description.

So what: more inference capacity outside Nvidia helps over time, but nothing here changes what you pay in 2026.

3. Meituan publishes LongCat-2.5 weights (9 November)

According to a summary of VentureBeat's report, Meituan published the weights of LongCat-2.5 under an MIT licence on Hugging Face on 9 November. Meituan's announcement says 1.6 trillion total parameters; the model card lists 1.8 trillion, a conflict unresolved at publication. Benchmark results are Meituan's own.

So what: open-weight models give you a self-hosting option, but a model this size needs serious infrastructure; price the hardware before getting excited.

4. TSMC approves up to $500 million each in Anthropic and OpenAI (10 November)

TSMC's filing with the SEC says its board approved investments of up to $500 million to buy Anthropic common shares at the per-share IPO price, and up to $500 million in OpenAI preferred stock. These are caps, not completed purchases. The same meeting approved about $40.9 billion in capital appropriations, mainly for advanced technology and packaging capacity.

So what: suppliers are buying stakes in their customers, which is the pattern I examine in the AI bubble debate post.

5. Australia announces an AI safety regulator plan (10 November)

The prime minister's release says the government will legislate mandatory AI safety and security protections for frontier AI companies operating in Australia, with a new regulator able to impose significant penalties. Companies would have to allow independent evaluation and report incidents quickly. Details come "in due course", with no date.

So what: this is a proposal aimed at model builders, not brands, but it adds to the pattern in my AI policy roundup.

6. Apple sells an Intelligence Usage Pack (announced 9 to 10 November)

9to5Mac reported that Apple's $49.99 one-time pack is its first paid Apple Intelligence tier. It works only inside Logic Pro for now, applies after monthly limits are reached, and is available in eight countries. Apple has not said how much extra usage it buys. The report expects extension to Final Cut Pro, Pixelmator Pro and iWork later this month; that is an expectation, not an Apple announcement.

So what: usage-based pricing for AI features is spreading, so expect more of your software to meter AI and budget accordingly.

7. Nvidia introduces GR00T 2 (10 November)

A summary of Nvidia's post says the robotics model averaged 71% success across 50 tasks with one demonstration per task. These are Nvidia's own results, and the same summary notes an independent evaluation reported 12.8% real-world success. It gives no public release date for the weights.

So what: nothing for a marketing team to do; read robotics benchmarks as vendor claims until independent tests arrive.

8. Anthropic: groups of personal agents do worse than one coordinator (11 November)

Anthropic's alignment team published simulated-environment results. When each user had their own agent, group outcomes were worse than with one coordinating agent, by about 10 to 45 percentage points across four environments. In the personal-assistant environment, Claude Opus 5 teams scored 44% of the best possible outcome against 88% for a coordinator. The authors caution that the environments were simulated and the scoring did not capture whether users consented to changes. An unread-message guard recovered 73% of failed episodes for Opus 5.

So what: if you let several people each run their own agent on a shared asset such as a content calendar, add a coordinator or a lock; see building a multi-agent content pipeline.

One item I could not verify: ByteDance's agent plan

The Next Web, citing Bloomberg, reported on 11 November that ByteDance will release a Muse-style AI agent within weeks, based on unnamed sources at an internal meeting. I did not open the underlying article, and the report gives no product name or date. So what: nothing to act on yet.

What this means for marketers and founders

My read of the week: the actionable item is OpenAI's business profiles, and everything else is context. This week I would do three things. Check whether your business qualifies for the beta and who owns the listing. Record a baseline of how ChatGPT describes you before changing any data. And decide which of your tools meter AI usage so a surprise bill does not land in January.

I would not change content plans because of TPUs, robotics models or funding stories. Revisit them when pricing or availability changes.

FAQ

What was the biggest AI news this week for marketers?

OpenAI's 10 November beta of Businesses in ChatGPT, which lets verified businesses manage a profile and see mention counts. Business Agents are limited to US e-commerce and online-service firms for now.

Does paying for ChatGPT ads change recommendations?

According to Digiday as cited by The Next Web, OpenAI says paying does not change how often ChatGPT recommends a business. I could not confirm that independently.

What are Google's TPU 8tx and 8i Max?

They are two new inference-oriented TPU variants announced on 9 November. Trade-press analysis says they are slated for 2027, and the pricing estimates are the author's speculation.

Is TSMC buying shares in Anthropic and OpenAI?

Its board approved investments of up to $500 million in each on 10 November, per its SEC filing. These are caps, and final purchase amounts were not disclosed.

Has Australia passed an AI safety law?

No. The government announced on 10 November that it will legislate and create a regulator, with details to follow and no date.

What did Anthropic's multi-agent research find?

In simulated environments, groups where each user had an agent did worse than groups served by one coordinator. The authors note the setups were simulated and that coordination does not scale easily.

Where do I find the Nvidia and pricing news?

I cover Nvidia's leadership change in this post and the compute-to-price link in this one.

Want help with this?

I'm Younus Fardeen, an organic growth, SEO and AEO strategist with 4+ years of marketing experience across edtech and startup teams. If you want help tracking how AI assistants describe your brand, or turning the weekly news into decisions, see my work and reach me through the contact form at younusfardeen.com.