Some comments from Scott Dawson, CEO at DECTA UK;
“The US government’s decision to prevent Claude Fable 5 from being released internationally is an interesting development for those of us who have been following the ongoing saga of the company’s ‘too powerful’ AI models.
This time, the issue is that there are ways to ‘jailbreak’ Fable 5, allowing users to leverage its frontier technology for purposes other than that which it was clearly intended. We’ve already seen a wave of AI-assisted fraud carried out from jailbroken AI models, such as China’s DeepSeek. Having a major AI model with 30 million active users able to be utilized for nefarious uses would be a disaster, so it is a cause for celebration that the US is being cautious. The price to pay is that the rest of the world won’t have access to frontier models as quickly, but it’s a price I’m personally willing to pay.
This situation highlights the ad-hoc way in which these far-reaching decisions are being made. It’s the nature of cybersecurity and IT in general that some of the biggest advances are made by the archetypal guy in his basement, and that seems to be what has happened here: an amateur found an exploit and reported it. An executive order has also been issued that prevents U.S states from creating their own AI rules, which leaves regulating AI to the Federal apparatus – in fact it was the Department of Commerce that suspended Fable.
The entire world needs real, robust regulatory policies and accountability in place that can inspire trust instead of confusion. What is needed is a much more systematized method for assessing and approving AI systems, and not just for cybersecurity but for other potential harms like ‘AI psychosis’. I won’t hold my breath for this given how little is currently understood by lawmakers about AI systems and the threat of fraud, but it is absolutely necessary.”
This is such a bs “article.” Amazon reported this as finding the “exploit”, not someone in their basement. Cyber security experts agree it’s not any more significant than any other model’s “exploits”. This is Trump admin retaliating for Anthropic refusing to weaponize their models to Pete Hegseth’s liking. This also gives other countries, who won’t be unilaterally shutting off their frontier models, to develop further ahead of ours. It also makes companies and people less likely to use US-based AI companies because of the instability of them being unilaterally shut off by the Trump admin.
This is a disastrous, unintelligent policy, and I’m sure this article was written by someone who in previous admins pulled the “those who seek security at the expense of liberty will get neither” every chance they got, to say nothing of wanting “smaller government”.
It’s a short opinion piece from DECTA UK’s CEO, free speech n all that…