A week in Generative AI: Glassing, Muse & Affordability

News for the week ending 12th April 2026

Apr 12, 2026

Three big news stories this week - Anthropic announced that they won’t release their newest model, Mythos, as it poses a major cybersecurity risk, Meta get back in the AI game, and Unitree announce pre-orders for their R1 robot outside of China for under $5k.

In Ethics News, a 20-year-old was arrested for allegedly throwing a Molotov cocktail at Sam Altman’s house, OpenAI shared their vision for the AI economy, and the US summoned bank bosses to discuss cyber risks posed by Anthropic’s latest AI model.

Good long reads from Helen Toner on AGI and the New Yorker on Sam Altman worth checking out too.

Anthropic debuts preview of powerful new AI model Mythos in new cybersecurity initiative

The frontier AI companies have been hinting at this for a while now - they’re going to get to the point where they train a model that has capabilities they’re hesitant to release to the general public. That moment has now come and the reason is cybersecurity.

This week Anthropic announced a new model, Mythos, that is so capable at coding that it poses serious cybersecurity risks. Instead of releasing it, they are partnering with some of the biggest technology companies in the world to test it and use it to identify security gaps in their own platforms before they decide whether to release Mythos more broadly. They’re effectively giving all the large technology platforms a ‘head start’.

To try and put this in context, Mythos found a bug that had gone undiscovered for 27 years. It’s also helped one cybersecurity professional discover more bugs in 2 weeks than they have throughout their entire career.

There is some scepticism that all of this is essentially ‘marketing cover’ to put more advanced models behind ever larger enterprise contracts but I don’t buy that argument. Although, I am a big cynical that complete co-incidence OpenAI also announced a few days later that they’re about release a new cybersecurity product.

What I think this shows is that we’re starting to reach a tipping point where models are so advanced that just releasing them isn’t a good idea. We’re now entering a period where frontier AI companies are going to have to think more carefully about how and when they release their most capable models.

Anthropic | TechCrunch | Axios

Meta debuts the Muse Spark model in a ‘ground-up overhaul’ of its AI

A month ago I wrote about Meta reportedly delaying the release of its latest model, but this week they did actually release it under the name Muse Spark. This is obviously Meta playing catch-up, so don’t expect it to introduce any new groundbreaking capabilities.

Simon Willison took it for a spin and noted that it has some interesting tools, and the release helped the Meta AI app climb to No. 5 on the App Store.

I’d file this as a stepping-stone release while Meta plays catch-up and they’ve promised that bigger more advanced models are to come and that they’re going to get back into the open-weights model game, which can only be a good thing.

Meta | TechCrunch | Simon Willison

Here’s The Most Affordable Humanoid Robot You Can Buy Now

Back in July last year, I shared that Chinese robotics company Unitree was releasing their R1 humanoid robot for less than $6k. They’ve just opened pre-orders for this outside of China for less than $5k.

It’s a bit mind-blowing to think that you could own a humanoid robot for less than $5k. It’s still a novelty at this point, and humanoid robots still can’t reliably complete your average day-to-day household chores but it’s looking increasingly likely that this is a solvable software problem, not a hardware problem now.

This means that it’s likely in the not too-distant future that people will be able to buy a genuinely helpful humanoid robot for less than $5k at today’s prices. If/when this takes off it’s going to have a massive impact on the world.

Forbes | Unitree

Web 4.0

AI Ethics News

Long Reads

Helen Toner - The term “AGI” is almost useless at this point
The New Yorker - Sam Altman May Control Our Future - Can He Be Trusted?

“The future is already here, it’s just not evenly distributed.“
William Gibson

Discussion about this post

Ready for more?