AI week: big promises, small releases and a security alarm

This week's AI news has a common theme: a lot is being launched and promised, but what is actually ready to use is often less than the headlines suggest.

The clearest example is Elon Musk's AI version of the Odyssey. In July, he promised a full-length feature film by the end of the year. Now a first episode of just over seven minutes has arrived. According to WION, reactions on X are mostly negative, though some praise the visuals. Musk's statement that Grok Bot will be able to call businesses for you and send a summary also remains undocumented, and pricing and rollout are unclear.

New models and lower prices

Among the big players, things are moving more concretely. Google has made Gemini 4 Argon available to paying subscribers, and the Gemini app can now use your Maps places and your calendar. Meanwhile, its successor Carbon is already being tested with select partners, where testers report better 3D but low speed.

Cheaper competition is coming from China. Moonshot AI's Kimi K3.1 Pro costs a third less than its predecessor and, according to Artificial Analysis, ranks fourth. That has many guessing that the anonymous coding model Kibitz on Arena is in fact Kimi, but this has not been confirmed. And according to an account linked to Chinese state TV, DeepSeek and MiniMax have been given the green light to use their AI abroad, while a draft of export controls is on the way.

Canva is taking a different route and has built its own image model, Craft, trained on the company's own templates. It is meant to make it easier to create posters and posts with text.

AI moves into the office

AI is increasingly becoming a regular part of work. OpenAI's ChatGPT Work is an agent, meaning a program that solves tasks on its own in multiple steps across work applications. Uber is now giving all roughly 30,000 employees access. Starting October 1, 2026, US government agencies can buy ChatGPT at a 50 percent discount and pay only for what they use. Meanwhile, ads in ChatGPT have expanded to 89 countries, a sign that the economics behind AI are still being worked out.

On the coding front, prices for AI programming help from Claude, OpenAI and Google can range from free to $500 a month.

The uncomfortable counterweight

When tools spread quickly, the risks follow. On October 21, version 1.107.0 of the popular tool LiteLLM was compromised with malicious code that steals credentials and gives the attacker remote access. It is a reminder that new AI tools often rest on many small building blocks, not all of which are equally safe.

What can you do?

Read launches with a critical eye: distinguish between what has been released and what has been promised. Many performance figures come from the companies themselves. If you or your workplace use AI tools, keep them updated and get a handle on who has access to what. Also keep an eye on whether Google's Carbon hits the market, and whether Musk's film ever gets finished.