A week in Generative AI: Fable 5, Siri AI & Robot Duck
News for the week ending 14th June 2026
Anthropic releases its first Mythos-class model Claude Fable, which then gets pulled by the US Government
This week has been a bit of a rollercoaster for Anthropic:
On Tuesday they launched Fable 5, the first publicly available Mythos-class model
On Thursday they then had to apologise for the invisible guardrails they had put in place on Fable 5
On Friday the US government ordered Anthropic to suspend access to both Fable 5 and Mythos citing national security concerns
On Saturday it was reported that it was Amazon that raised the national security concerns with the US Government
So, yes, there is a lot to cover.
The capabilities of Fable 5 are clearly part of the story. Based on the benchmarks and early reports, it’s the most capable model released so far. It can handle more complex tasks and run for much longer without human intervention.
The invisible guardrails were a mistake. I can understand why Anthropic wanted to prevent people using Fable 5 to train rival models, but users should always know when a model is being restricted, redirected or degraded.
Then there is the Amazon angle, which makes the whole thing even stranger. Amazon is an Anthropic investor. It is also a customer, distribution partner and competitor. So if Amazon really did raise concerns with the US government, it shows how messy the market dynamics around frontier AI have become.
This all boils down to the bigger issue: regulation.
If the US government believes Fable 5 and Mythos present a national security risk, then that risk is unlikely to be unique to Anthropic. In fact, Anthropic says other models are capable of similar narrow jailbreaks.
So why, three and a half years after ChatGPT launched, do governments still not have a clear framework for how to regulate frontier AI models?
And why is there still so little focus on the increasingly entangled relationships between the big AI players?
If you’re interested in some reviews of Mythos and Fable, both Ethan Mollick and Simon Willison have some thoughts worth reading:
Ethan Mollick - What it feels like to work with Mythos
Simon Willison - Initial impressions of Claude Fable 5
YouTube | Anthropic | The Verge
Apple wants Europe to blink
As Apple announced at WWDC on Monday, Siri AI will be delayed in the EU due to DMA. This is a different kind of regulatory challenge that we’ve seen for AI-adoption in the EU as it’s not actually about the capabilities of the underlying AI model.
The issue is much more fundamental - the EU classifies Apple and the iPhone as a ‘gatekeeper’ and because of that they insist that EU citizens have the right to choose what features to use on their devices. This is relatively straightforward when it comes to things like web browsers, email apps etc. but gets very complicated very quickly when it comes to things like app stores and AI assistants.
The EU’s argument is that for Apple to be able to launch Siri AI in the EU that they will need to let other AI assistants on iPhones have the same level of access to all the information on the device. Apple’s argument is that this would require them making devices less secure as Siri AI works with access to a huge amount of personal data on device.
Unfortunately for EU citizens this looks like it’s going to remain as an impasse and because of that I’m not expecting Siri AI to launch in the EU at all as things stand, which is very disappointing.
There has however been lots more coverage of Siri AI since I published my thoughts on Tuesday that are worth checking out if you want to get a better idea of what it’s capable of. General reviews have been very encouraging!
Robot Duck
I’ll be keeping an eye on this little project as it looks fun! Google’s Gemma 4 model is their latest small, on device model and they’ve put that into a small bi-pedal robot to see what they can do with it.
We’re used to having a chat interface for interacting with large language models, but voice is an obvious natural fit and lends itself well to being the main way of interacting with a robot.
The next step is getting Gemma 4 to be able to control the robots - I’m looking forward to seeing how they get on!
AI Ethics News
White House Defangs AI-Testing Unit at the Worst Possible Time
World’s first wind-powered underwater datacentre starts operating in China
Microsoft AI chief on why it’s ‘dangerous’ to call AI ‘alive’
Long Reads
Stratechery - The iPhone’s Last Stand
OpenAI - Built to benefit everyone: our plan
“The future is already here, it’s just not evenly distributed.“
William Gibson