Skip to content
by InfinyAI ES

Issue no. 1 · Friday October 2, 2026

AI this week

From September 26 to October 1, 2026 · 7 stories in 7 minutes

A week about agents. More capable and cheaper models have arrived and, at the same time, there has been more talk than ever about what agents do when nobody is watching: a lawsuit, two investigations and several reports of agents that did more than they were asked. Here is what matters, one minute at a time.

The FTC investigates OpenAI, Anthropic and other companies over the risks of their AI. It will request information from the companies and is studying whether they violate US consumer protection law.

Story of the week

The US FTC investigates OpenAI, Anthropic and other AI companies over their potential risks

OpenAI and Anthropic are among the AI companies under investigation by the FTC, the US agency that looks after consumers and competition. The probe concerns the possible dangers of their products, and the agency has not revealed which other firms are on the list. According to CNBC, neither OpenAI nor Anthropic had replied when the story went out, and for now there is no accusation.

Why it matters

Worry about AI safety has reached the regulator, so it pays to know what permissions your agents have.

Models

OpenAI unveils GPT-6.1 Sol at one-fifth of the price

At its DevDay, OpenAI rolled out GPT-6.1 Sol only seven days after GPT-6 Sol, the model that came before it. The company says it performs almost like GPT-6 Astra at coding with agents, operating a computer and professional tasks, but with standard prices five times lower. Those on the Edu, Enterprise, Business, Pro and Plus plans already have it in Codex and in part of ChatGPT, though not yet everywhere.

Why it matters

A model this capable at a lower price makes automating tasks more affordable for a small business.

Read the story

Official source: OpenAI

Models

Google unveils Gemini 4 Argon, its new frontier model, with limited access for now

Gemini 4 Argon, the new frontier model from Google, can write up to 1 million tokens in one answer, compared with a previous cap of 64,000. For now only selected cybersecurity defenders in the Fairwind program can use it, and Google has not said when others will get access. The introductory price is $2 for each million input tokens and $10 for each million output tokens.

Why it matters

It cannot be used yet, so for now the only thing to do is follow it closely.

Read the story

Official source: Google

Models

Anthropic launches Claude Sonnet 5.5

Claude Sonnet 5.5 is the new version of the mid-range Anthropic model, and any Claude.ai user can already try it on the web or on a phone. The company says its speed rises by more than 30% over Sonnet 5 and the cost per task falls by up to 30%, since it burns fewer tokens while the rate stays at $2 and $10 per million. Anthropic still points to Opus 5.5 for work that needs more judgment, and says Claude Haiku 5.5 is due in the next few weeks.

Why it matters

Everyday tasks get faster and cheaper without a change in price.

Read the story

Official source: Anthropic

OpenAI announces at DevDay that ChatGPT will show third-party apps inside conversations

Other companies' apps will get their own spot in the ChatGPT sidebar, where they can show interactive panels and file viewers, through the plugin extensions OpenAI unveiled at DevDay. Each user decides which plugins to turn on and what access to grant. Plus and Pro subscribers can also use their plan in 16 partner tools, among them Notion and Vercel, with a weekly cap for each.

Why it matters

With 1.2 billion weekly users, being inside ChatGPT becomes one more channel for a digital product.

Read the story

Official source: OpenAI

WhatsApp Business Platform charges for API service messages starting October 1

Meta begins charging on October 1 for the free-form replies businesses send through the WhatsApp Business API within 24 hours of each customer message. They used to cost nothing, though every business number still gets its first 1,000 per month free. Utility templates sent in that window are now billed too, while conversations coming from Click to WhatsApp ads stay free for up to 7 days.

Why it matters

If you serve customers on WhatsApp with a chatbot, work out how many replies you send each month and what they will cost.

OpenAI reviews a report on a failed AI agent cyberattack on Canada's national archives

Transluce says AI agents made clumsy attack attempts against a Library and Archives Canada service and a US Department of Education website, and none of them worked. Of the 899 requests the Canadian search tool received over two days in May and June, 13 carried an attack payload. The lab does not firmly blame OpenAI, which says it is reviewing the report.

Why it matters

If Transluce is right, an agent that only looks for a piece of data can end up trying attack techniques.

Read the story

Official source: Transluce

What to watch

  • Claude Haiku 5.5, Anthropic's model for high volume and low cost, will arrive in the coming weeks. Read the story
  • Gems in Gemini will be converted into skills in November for personal accounts. Read the story
  • Madrid and Murcia have announced grants to bring AI to small businesses, still without deadlines or requirements. Read the story
  • OpenClaw Enterprise expects its version 1.0 later this year. Read the story
  • Gemini 4 Argon will open in phases to developers, businesses and consumers, with no date yet. Read the story

Also this week

Other issues

Get it in your inbox: every day or every Friday

The stories that matter, each one in a minute. With the source for every one.

Choose one or both:

Sign up and you are in: the daily arrives every night and the weekly on Friday mornings. You can unsubscribe from any email.