Updated 12 November 2026. AI moved fast again this month. In just over a week we got a new flagship model from Anthropic, a tiny agent model from Meta, a "kill switch" for rogue AI agents and the run-up to the biggest AI IPO so far. Here are the seven AI trends founders, CTOs and developers are talking about right now, with what each one means for your business.
Looking for the earlier headlines? Read our AI news roundup from 3 November.
The short version
- Frontier AI got faster and cheaper: Anthropic's Claude Fable 5.2 is up to 30% faster and 20% cheaper than the model it replaces.
- Tiny "worker" models are here: Meta's Muse Ember costs 10 cents per million input tokens and is built for AI agents.
- Cyber defenders get the strongest models first: Google's Gemini 4 Carbon launched for security teams before anyone else.
- AI agents now come with an off switch: CrowdStrike launched an "AI Kill Switch" to stop agents that go off script.
- Agents are becoming always-on: OpenAI's Dots work in the background on their own cloud computers.
- AI is now a trillion-dollar business: Anthropic started the roadshow for its IPO.
- Your AI supplier can change the rules: OpenAI's contract to supply models to the coding tool Cursor is due to end today.
1. Frontier models are getting faster and cheaper
On 9 November, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, its most capable models. According to CNBC, they produce answers about 30% faster than the 5.1 generation and cost 20% less: $8 per million input tokens and $40 per million output tokens.
Fable 5.2 is available to everyone. Mythos 5.2 stays limited to vetted organisations in Anthropic's life sciences and cyber verification programmes. Anthropic pitches Fable 5.2 at hard engineering, mathematics and robotics problems, and suggests its September model, Claude Opus 5.5, for everyday work.
What it means for you: the same AI feature costs less to run every few months. If you shelved an AI idea last year because it was too expensive, it is worth costing it again. And match the model to the job: the most powerful model is rarely the cheapest way to answer a customer email.
2. Tiny "worker" models built for AI agents
Also on 9 November, Meta released Muse Ember, which it calls the lightest and most cost-efficient model from Meta Superintelligence Labs. TechCrunch reports a price of 10 cents per million input tokens. Ember understands text, images and audio, so an agent can read screenshots, charts and scanned documents with it.
Meta pairs it with its larger Muse Wonder model: "Muse Wonder plans, Muse Ember handles the volume." Meta also claims Ember beats rival small models on cost and quality. Those are the company's own numbers, so test before you trust them.
What it means for you: the smart way to build an AI agent in 2026 is a planner and a worker. A strong model decides how to handle a task, and a cheap, fast model does the repetitive work: sorting tickets, reading invoices, filling forms. This can cut running costs a lot for high-volume jobs.
3. The most powerful models go to cyber defenders first
Google announced Gemini 4 Carbon on 28 October, two weeks after Gemini 4 Argon. 9to5Google reports that Google ranks it first on the Artificial Analysis Intelligence Index v5.0, and that it went to "cyber defenders" first. It is "coming soon" to developers, businesses and consumers, with no date yet. The introductory price is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 later.
Anthropic is doing something similar: Mythos 5.2 is only for vetted security and science organisations. The reason is simple. These models are now very good at finding and fixing software vulnerabilities, and the companies want defenders to get them before attackers do.
What it means for you: attackers will use AI to find weak spots too. Keep your software and its libraries up to date, and get a security review of anything that handles customer data or payments. And don't plan a product around a model you can't use yet.
4. "Kill switches" for AI agents
On 3 November, CrowdStrike launched an AI Kill Switch in its Falcon Guardian product. It lets an authorised administrator stop an AI agent that is acting outside its intended scope, prevent it from relaunching, cut off its network access, and keep an audit trail of every action. For now it stops agents at the gateway; stopping them directly on devices is coming later.
What it means for you: as soon as an AI agent can send emails, change records or spend money, you need a way to stop it. Every agent you deploy should have limited permissions, a log of what it did, human approval for risky actions and an off switch. These are much easier to build in at the start than to add later.
5. Always-on AI agents with their own computers
OpenAI's Dots, announced at its DevDay conference and now rolling out, are agents inside ChatGPT that keep working when you are not in the chat. According to SiliconANGLE, each Dot runs on its own cloud computer that you can open to watch, connects to more than 4,000 apps through plugins, and can be reached in ChatGPT, Slack and Microsoft Teams. Actions that touch accounts or share information go through an "auto-review" check, and OpenAI warns the agents "can still make mistakes".
Dots are rolling out to ChatGPT Pro and Business Premium subscribers, with an enterprise beta. They are not yet offered to Pro subscribers in the EEA, Switzerland or the UK.
What it means for you: customers and staff will soon expect software that an AI assistant can work with. If your product or internal tools have clean APIs and clear permissions, agents like these can use them. If they don't, now is a good time to fix that.
6. The trillion-dollar AI IPO
Anthropic filed for its IPO on 26 October and launched its roadshow on 10 November. CNBC reports the company is aiming for a valuation of $1 trillion to $1.5 trillion, down from a $2 trillion figure floated in recent months, and that it reported $33.24 billion in revenue for the nine months to 30 September, up about 1,200% on a year earlier.
What it means for you: AI is no longer an experiment for the companies that make it. It is a very large business with very large customers. Expect the big AI providers to focus on business features, reliability and pricing plans, and expect public-market pressure to shape their prices and products.
7. Your AI supplier can change the rules
In August, OpenAI said it would stop supplying its models to Cursor, the AI coding tool, after SpaceX bought it. The planned cut-off date is today, 12 November. Engadget reports that OpenAI said it could not be confident SpaceX would follow its terms of service. Cursor's CEO said OpenAI served about 5% of Cursor's customers, and Anthropic said it would keep supporting Claude in Cursor.
What it means for you: if your product depends on one AI company, that company's business decisions become your risk. Build your AI features so the model can be swapped: keep prompts and logic in your own code, put the AI provider behind a single connector, and test a second provider before you need it.
What this means for your business
- Start with one workflow. Pick a repetitive, high-volume task, such as support tickets, invoices or lead qualification, and automate that first.
- Mix big and small models. Use a strong model to plan and a cheap one to do the volume work.
- Build safety in from day one. Limited permissions, logs, human approval for risky steps and an off switch.
- Stay vendor-neutral. Make it easy to switch AI providers when prices, terms or quality change.
- Tighten your security. AI makes finding vulnerabilities faster for everyone, including attackers.
Frequently asked questions
What are the biggest AI trends in November 2026?
Faster and cheaper frontier models (Claude Fable 5.2), small models built for AI agents (Meta Muse Ember), security-first launches (Gemini 4 Carbon), kill switches for AI agents, always-on agents such as OpenAI's Dots, Anthropic's IPO, and the risk of depending on a single AI provider.
What is Claude Fable 5.2?
Anthropic's most capable generally available model, released on 9 November 2026. It is about 30% faster than the previous version and costs $8 per million input tokens and $40 per million output tokens, 20% less than before.
What is Meta Muse Ember?
A lightweight, low-cost model from Meta, released on 9 November 2026 and designed to do the repetitive work inside AI agents. It costs 10 cents per million input tokens and can read text, images and audio.
What is an AI kill switch?
A control that lets an administrator stop an AI agent immediately, stop it from restarting and cut off its access when it behaves outside its intended scope. CrowdStrike launched one on 3 November 2026.
Can I use Gemini 4 Carbon yet?
Not generally. Google gave cyber defenders access first and says it is coming soon to developers, businesses and consumers, without a date yet. Gemini 4 Argon is available now.
How can a small business use AI agents safely?
Start with one well-defined task, give the agent only the access it needs, log everything it does, ask for human approval before risky actions such as payments or deletions, and make sure someone can switch it off.
How Bitinfusion can help
We design and build AI automation and agents and AI chatbots for businesses in India and around the world. We build them with the guardrails above, and so you can switch AI providers when you need to. Working from another country? See how we work with clients abroad.
Have a workflow you would like to automate? Book a free call at a time that suits you.
Sources
- Anthropic newsroom (Claude Fable 5.2 and Mythos 5.2, 9 November; IPO roadshow, 10 November)
- CNBC: Anthropic releases Claude Fable 5.2 and Mythos 5.2
- TechCrunch: Meta releases Muse Ember
- 9to5Google: Gemini 4 Carbon
- CrowdStrike: AI Kill Switch press release
- SiliconANGLE: OpenAI launches Dots
- Engadget: OpenAI will pull its models from Cursor