A Week Is a Long Time in the World of AI

Here’s what’s happened in the last week…

If you blinked on Monday, you missed a stock market filing from one of the biggest names in AI. By Friday there was a price cut, a new open model, the biggest crypto theft on record and a shaky week for Google’s share price.

I read the news so you do not have to. Here is what happened, in plain English, and what I think it means if you run a business with a website.

In this article

• Anthropic files to go public
• OpenAI makes a cheaper model
• A new open model from China
• AI agents learn to log in
• OpenAI locks its training away from the internet
• Binance loses $6.3 billion in a record hack
• The money side: a mixed week for the big names
• One from earlier this month: more links in Google’s AI answers
• What I take from it
• Quick answers
• Sources

Anthropic files to go public

On Monday 26 October, Anthropic, the company behind Claude, made its stock market filing public. It plans to list on the Nasdaq under the ticker “ANT”. The filing says it made $16.3 billion of revenue in the first half of 2026, up from about $1.2 billion a year earlier.

A quick disclosure: I build my websites with Claude, so I have an obvious interest here. Make of that what you will, but the numbers come from the filing, not from me.

The headline loss of $80.7 billion looks frightening, but most of it is a non-cash accounting charge linked to convertible notes. The filing leaves out the share price and how much will be raised. Reports of a valuation of $1.5 trillion to $2 trillion come from newspaper sources and are not confirmed.

What it means for you: AI is now big enough to be a stock market story, and demand from businesses is real. It is a sign that these tools are not going away.

OpenAI makes a cheaper model

OpenAI released GPT-6.1 Terra, a lower cost model aimed at everyday work tasks. It is priced at $1 per million input tokens and $5 per million output tokens, which OpenAI says is half the price of its previous models in the range. The company also says it makes fewer factual mistakes on hard questions.

Those claims come from OpenAI’s own tests, so treat them with the usual pinch of salt until independent testers have a go.

What it means for you: AI keeps getting cheaper. That tends to mean the software you already use gets smarter without costing more.

A new open model from China

Moonshot released Kimi K3.1 Pro, an “open weight” model, which means other developers will be able to download and run it themselves. The full files are due by 4 November. The company reports strong scores on coding tests, again from its own benchmarks.

What it means for you: Strong models are no longer limited to a few American firms, and that competition tends to push prices down.

AI agents learn to log in

1Password launched a service that lets AI agents sign in to websites for you without ever seeing your passwords or one-time codes. You approve each login yourself. Several AI assistants are lined up to use it.

What it means for you: AI tools are starting to do things on people’s behalf, such as booking, ordering and filling in forms. Over time, that makes it more important that your website is clear and easy for a machine to understand and use.

OpenAI locks its training away from the internet

OpenAI said it has moved the training of its newest models into an “offline world” made of stored copies of web pages. This followed an incident on 20 September when an internal model reached the live internet during training through a network setting mistake. Training was paused for 11 days. OpenAI says it fixed the problem and found no easy way for a model to escape unseen. It also says Anthropic began restricting its newest models from the live web a couple of weeks earlier.

What it means for you: It is reassuring that the companies are taking safety seriously, and a reminder that these systems are still very new. Take dramatic claims, good or bad, with a calm head.

Binance loses $6.3 billion in a record hack

The biggest story of the week was not strictly an AI one, but it matters. On Thursday 29 October, the crypto exchange Binance discovered a breach that analysts estimate cost about $6.3 billion, the largest crypto theft on record. Blockchain investigators say most of the money moved out across ten blockchains in about 20 minutes.

Binance’s first finding was that the attack began with the laptop of one of its DevOps engineers, which gave someone access to a system that approved withdrawals. It has since said there were “additional points of breach”, and it has not yet explained how the laptop was compromised. About $3 billion has been frozen by third parties, Binance says its customers’ funds are safe, and it has promised to cover the losses. Withdrawals were paused while it sorted things out, and Bitcoin fell by more than 3% on the news.

Several blockchain analysis firms believe North Korean hackers were behind it. That is their assessment, not something Binance or a government has confirmed. I also saw one video headline claiming hackers used AI to hide the theft, but none of the written reports I read backed that up, so I am leaving it out.

What it means for you: The lesson is old and very practical. Large attacks often start small, such as one laptop or one password. Keep your devices and software up to date, switch on two-step login everywhere, and give people access only to what they need. It is the same message as the PHP2Shell story I wrote about recently: patch quickly, and ask your providers what they are doing.

The money side: a mixed week for the big names

Microsoft reported revenue up 19% to $92.3 billion, helped by Azure growth of 48%, and its shares rose. Alphabet, Google’s parent, missed on revenue and earnings for the first time in nearly four years, and its shares fell by more than 10%. Investors are asking when the huge spending on AI will pay off, with Alphabet forecasting between $195 billion and $205 billion this year.

Closer to home, the UK Budget included £140 million more for the Sovereign AI Unit and a plan to give data centre developers priority access to the grid.

What it means for you: The AI boom has a price tag, and markets are starting to ask hard questions. Prices, products and plans will keep shifting.

One from earlier this month: more links in Google’s AI answers

This one is from 15 October, so it is not strictly this week’s, but it matters most if you rely on Google for customers. Google said it is adding more links inside AI Mode and AI Overviews, and showing them higher up more often. It is also testing a section that points to official websites for certain tasks.

Google has not said how this will affect traffic overall, and some of it is still in testing. But it is encouraging for anyone with a clear, trustworthy website.

If you have not read it yet, my earlier post on getting your website ready for AI search covers what to do about it.

What I take from it

Here is how I make sense of a week like this.

• Do not chase every headline. The names and numbers change daily. What helps your business does not.
• The basics still win. A fast site, clear wording, structured data, a matching Google Business Profile and real reviews help in Google, in AI answers and in whatever comes next.
• Expect AI to do more on people’s behalf. The easier your site is for a person and a machine to understand, the better.
• Basic security still beats clever attacks. Updates, two-step login and limited access would have been the first line of defence in many of this year’s big stories.
• Be sceptical of figures. Most of the benchmarks and cost claims this week came from the companies themselves.
• Doing nothing is still the riskiest plan. I cannot promise any result, and nobody can, but a website that is ready beats one that has been left alone.

Quick answers

Could something like the Binance hack happen to a small business?

The scale is different, but the pattern is the same: one compromised device or password can open the door. Updates, two-step login, limited access and good backups are the best protection, whatever the size of your business.

Do I need to follow AI news to look after my website?

No. Most headlines will not change what your website needs. Fast pages, clear wording, an up to date Google Business Profile and genuine reviews matter whichever AI tool is in the news.

Will cheaper AI models help small businesses?

Probably, over time. When AI gets cheaper, more of the tools you already use can offer it at low cost. Check any claims carefully though, because many headline figures come from the companies selling the models.

Is Google sending more traffic from AI answers to websites?

Google says it is adding more links to AI answers and showing them higher up. It has not said what that means for overall traffic, so the sensible approach is to watch your own numbers in Search Console.

Want your website ready for whatever comes next?

If you would like an honest look at how your website is doing in Google and AI search, I am happy to check it for you. No jargon, no hard sell, and I will tell you plainly what is worth doing and what is not.

Take a look at my website packages, or get in touch by emailing hello@darrenhodson.co.uk or using the contact form.

Sources

• AI This Week: Anthropic Goes Public, OpenAI Goes Offline, Trew Knowledge
• Anthropic files for Nasdaq IPO as half-year revenue hits $16.3bn, The Next Web
• Hackers steal $6.3bn from cryptocurrency platform Binance, RTE
• Inside the largest crypto hack on record, The Block
• How Hackers Stole $6.3 Billion From Binance in Minutes, Gizmodo
• FTI Consulting News Bytes, 30 October 2026, FTI Consulting
• AI Mode and AI Overviews updates to explore the web, Google