Jennifer Holly is a penultimate year Classics and English student at Magdalen College, Oxford.
The news coming out of America this week has been deeply worrying for the tech sphere, but its severity demands that anyone interested in the future of our society must listen carefully.
Anthropic, the AI company behind Claude, had been, since April, teasing a new model – Mythos – that was purportedly such a risk to cybersecurity that it could not be released to the public. Last week, it released Fable 5, a model built on Mythos’s infrastructure, but crucially protected with guardrails that prevent its use for society-threatening ends, such as devastating cybersecurity hacks. The US government quickly became anxious of a “jailbreak” that allowed these guardrails to be surpassed, and imposed export controls on Mythos and Fable, meaning that only US citizens will be able to access them. A truce is currently in negotiation between Anthropic and the US Government.
There is much speculation over the motives of the US government and Anthropic in this scenario. By way of brief background – the US Department of War and Anthropic have been involved in a feud since January, as the latter, who have branded themselves as safety-focused, refused to give the DoW unfettered access to use Claude models for mass surveillance and autonomous drone strikes. Other elements of the US administration, then, are baffled that Anthropic is seemingly unbothered by what they believe is a critical breach of safety. They have stated that these two negotiations (or lack thereof) are entirely disparate – but what emerges is a picture of AI deeply embroiled in the security, sovereignty, and subsequent prosperity of nations.
Keir Starmer went grovelling to Trump to secure a spot on the list of countries with approved access to frontier AI models, but was met with a snub. What this direly exposes is the sheer extent of Britain’s AI insecurity. However this particular scenario unfolds, it is imperative that AI is treated as a key national asset – and as with anything worth keeping, its owners will be all too happy to pull it away to their own advantage.
If we factor this threat into the picture, here’s how it could look:
The US currently controls around 75 per cent of global AI compute (the hardware and energy that allows AI to run) – and Britain controls a paltry 3 per cent. Britain, undoubtedly, boasts good innovation in AI, but the vast majority of British startups – and the core of its financial services sector – are not running on their own hardware, but are reliant upon US cloud providers, such as AWS and Azure, and frontier models from US companies, namely OpenAI, Anthropic, and Google.
If AI is rationed – say, if the US divvies out a limited amount of tokens (which govern the reasoning behind LLM answers) to other nations, there will be little hope for any continued success in British AI innovation, and – to be frank – the entire basis of Britain’s economy will be threatened.
If Starmer was under the belief that he could curry the US administration’s favour, he may have looked at the paradox of Britain’s AI Security Institute as a sobering example. The AISI – rather oddly for a public body – is highly regarded within the AI world, and was recruited to test the safety of Fable.
They made “substantial progress towards a universal jailbreak”, which essentially means that a malevolent actor could, potentially, harness Fable’s power beyond its guardrails. The AISI, despite their expertise, will now be locked out of Anthropic’s frontier models altogether by the US’s hand – leaving top engineers in limbo, simply because they are not US citizens. For the AISI and for everyone in Britain, the only solution is to build domestic AI power.
The more the British government makes a failure of this, the more it makes a fool of itself.
Britain’s AI sovereignty is currently being blocked, as the well-worn adage tells, by planning backlogs and sky-high energy prices. These same, repetitive, systemic issues can be applied across Britain’s ailing economy – but AI demands particular concern, as it progresses with a rapidity that simply cannot wait for a planning committee’s approval.
On an individual level, it is deeply understandable to be sceptical about AI. It produces slop, offloads our cognition, and (paradoxically to the slop) threatens our way of life – which seems to be on the US government’s mind. But whatever we may make of it in our everyday lives, it cannot be swept under the rug that it is actively revolutionising military activity, and it would be negligent to dismiss its potential to take our economy into a new age.
To be locked out of this is to risk sinking into irrelevance, poverty, or vassalage.
What is required is a dismantling of heuristics very much entrenched – Britain cannot rest on her laurels as a lauded “surveyor” of AI, when in fact she owns none of it. The only strategy that can rescue her is the opening of a cascade of capital into Britain’s AI development – which necessitates the treatment of AI as of prime importance to national security.
For our future, I can only hope that we awake to this soon.
Jennifer Holly is a penultimate year Classics and English student at Magdalen College, Oxford.
The news coming out of America this week has been deeply worrying for the tech sphere, but its severity demands that anyone interested in the future of our society must listen carefully.
Anthropic, the AI company behind Claude, had been, since April, teasing a new model – Mythos – that was purportedly such a risk to cybersecurity that it could not be released to the public. Last week, it released Fable 5, a model built on Mythos’s infrastructure, but crucially protected with guardrails that prevent its use for society-threatening ends, such as devastating cybersecurity hacks. The US government quickly became anxious of a “jailbreak” that allowed these guardrails to be surpassed, and imposed export controls on Mythos and Fable, meaning that only US citizens will be able to access them. A truce is currently in negotiation between Anthropic and the US Government.
There is much speculation over the motives of the US government and Anthropic in this scenario. By way of brief background – the US Department of War and Anthropic have been involved in a feud since January, as the latter, who have branded themselves as safety-focused, refused to give the DoW unfettered access to use Claude models for mass surveillance and autonomous drone strikes. Other elements of the US administration, then, are baffled that Anthropic is seemingly unbothered by what they believe is a critical breach of safety. They have stated that these two negotiations (or lack thereof) are entirely disparate – but what emerges is a picture of AI deeply embroiled in the security, sovereignty, and subsequent prosperity of nations.
Keir Starmer went grovelling to Trump to secure a spot on the list of countries with approved access to frontier AI models, but was met with a snub. What this direly exposes is the sheer extent of Britain’s AI insecurity. However this particular scenario unfolds, it is imperative that AI is treated as a key national asset – and as with anything worth keeping, its owners will be all too happy to pull it away to their own advantage.
If we factor this threat into the picture, here’s how it could look:
The US currently controls around 75 per cent of global AI compute (the hardware and energy that allows AI to run) – and Britain controls a paltry 3 per cent. Britain, undoubtedly, boasts good innovation in AI, but the vast majority of British startups – and the core of its financial services sector – are not running on their own hardware, but are reliant upon US cloud providers, such as AWS and Azure, and frontier models from US companies, namely OpenAI, Anthropic, and Google.
If AI is rationed – say, if the US divvies out a limited amount of tokens (which govern the reasoning behind LLM answers) to other nations, there will be little hope for any continued success in British AI innovation, and – to be frank – the entire basis of Britain’s economy will be threatened.
If Starmer was under the belief that he could curry the US administration’s favour, he may have looked at the paradox of Britain’s AI Security Institute as a sobering example. The AISI – rather oddly for a public body – is highly regarded within the AI world, and was recruited to test the safety of Fable.
They made “substantial progress towards a universal jailbreak”, which essentially means that a malevolent actor could, potentially, harness Fable’s power beyond its guardrails. The AISI, despite their expertise, will now be locked out of Anthropic’s frontier models altogether by the US’s hand – leaving top engineers in limbo, simply because they are not US citizens. For the AISI and for everyone in Britain, the only solution is to build domestic AI power.
The more the British government makes a failure of this, the more it makes a fool of itself.
Britain’s AI sovereignty is currently being blocked, as the well-worn adage tells, by planning backlogs and sky-high energy prices. These same, repetitive, systemic issues can be applied across Britain’s ailing economy – but AI demands particular concern, as it progresses with a rapidity that simply cannot wait for a planning committee’s approval.
On an individual level, it is deeply understandable to be sceptical about AI. It produces slop, offloads our cognition, and (paradoxically to the slop) threatens our way of life – which seems to be on the US government’s mind. But whatever we may make of it in our everyday lives, it cannot be swept under the rug that it is actively revolutionising military activity, and it would be negligent to dismiss its potential to take our economy into a new age.
To be locked out of this is to risk sinking into irrelevance, poverty, or vassalage.
What is required is a dismantling of heuristics very much entrenched – Britain cannot rest on her laurels as a lauded “surveyor” of AI, when in fact she owns none of it. The only strategy that can rescue her is the opening of a cascade of capital into Britain’s AI development – which necessitates the treatment of AI as of prime importance to national security.
For our future, I can only hope that we awake to this soon.