Anthropic has been pressured to droop two of its synthetic intelligence (AI) fashions globally after the US authorities issued a directive prohibiting their use by overseas nationals over “nationwide safety” issues.
The transfer to droop Anthropic’s fashions got here simply days after the launch of the Claude Mythos 5 frontier mannequin, with the US authorities claiming it had “develop into conscious” of a jailbreaking methodology that allowed its safeguards to be bypassed.
Entry to Mythos 5 has been restricted by the agency and is just accessible by means of a “trusted entry program” of round 50 organisations – together with cyber safety companions and researchers – that use it to conduct superior cyber safety evaluation to search out and patch vulnerabilities earlier than the expertise is accessible to attackers.
Fable 5 is the consumer-facing model constructed with further safeguards, which the US authorities discovered a way of jailbreaking.
Anthropic made clear in its assertion that it didn’t agree with the US authorities’s resolution to recall the AI fashions. It pushed again on claims of the fashions’ vulnerability, describing it as “minor” and “comparatively easy”, including that “if this normal was utilized throughout the business, we consider it will primarily halt all new mannequin deployments for all frontier mannequin suppliers”.
That is the primary time the US authorities has exercised an export management directive towards an AI agency. In response, greater than 80 cyber safety leaders, together with Nvidia and Adobe, signed an open letter in assist of Anthropic, urging the federal government to elevate the management directive and guarantee rules are “enforced transparently and pretty”.
Rising issues over UK’s reliance on US tech
The episode between Anthropic and the US authorities comes amid growing concern amongst UK MPs concerning the over-reliance on US tech firms, which has beforehand led to requires the UK authorities to again sovereign IT.
Following the suspension, Baroness Beeban Kidron questioned the federal government within the Home of Lords about whether or not the elevated dependence on US firms in well being, schooling and safety created a crucial vulnerability for nationwide safety.
Lots of right this moment’s technological dependencies have been established in an period of relative geopolitical stability and a baseline assumption of belief. Latest occasions, nonetheless, show that this assumption has been undermined Sharinee Jagtiani, German Marshall Fund
Responding to Kidron on behalf of the federal government, Baroness Lloyd of Effra didn’t straight deal with the issues raised about nationwide safety, however did stress the necessity for the UK to develop sovereign AI capabilities. “Our strategy is about constructing energy over key elements of the worth chain to carry to the desk applied sciences that nobody else can do with out,” she stated.
Issues over the UK’s reliance on US expertise have been additionally raised by MPs in April 2026, after a report by the Open Rights Group (ORG) flagged the nationwide safety and financial dangers to the UK’s crucial infrastructure, and the potential of providers being shut down by US order.
ORG beforehand warned, in January 2026, that hyperscale cloud providers reminiscent of Microsoft Azure and Amazon Net Providers (AWS) are certain by US legislation, that means American authorities maintain the ability to demand entry to information saved within the UK.
Sharinee Jagtiani, a geopolitical analyst on the US-based German Marshall Fund, stated: “Lots of right this moment’s technological dependencies have been established in an period of relative geopolitical stability and a baseline assumption of belief. Latest occasions, nonetheless, show that this assumption has been undermined.”
The subsequent step ought to be targeted on mitigating threat and shifting in the direction of “managed interdependence” to extend resilience, stated Jagtiani. European international locations are prone to begin by advancing coverage initiatives to assist home expertise ecosystems and cut back publicity to exterior shocks.
“International companies and public establishments should now function in an surroundings the place technological dependencies might be weaponised, and issue that in as they construct out their enterprise fashions and insurance policies,” Jagtiani added.
This consists of strengthening a rustic’s function in rising expertise chains, growing a various provider base slightly than relying too closely on a single companion – for instance, by deepening connections with “expertise center powers” reminiscent of Canada, Australia, Japan, South Korea and India.
Anthropic versus the US authorities
Tensions between Anthropic and the US authorities have been escalating for a while. This started when Anthropic’s CEO, Dario Amodei, denied the US navy use of its AI fashions for “absolutely autonomous deadly weapons or mass home surveillance of Individuals” in February 2026.
The US authorities responded by placing Anthropic on a nationwide safety blacklist – with defence secretary Pete Hegseth labelling the corporate a “provide chain threat” – that means the service isn’t safe sufficient for presidency use. Anthropic sued the US Division of Defence over this transfer.
Claude was prohibited from use within the Pentagon consequently. Regardless of this, the AI mannequin was utilized by the US navy within the battle in Iran, as reported by CNBC.
A US federal decide finally dominated that the Pentagon’s directive couldn’t be enforced, discovering that the federal government’s measures appeared designed to punish Anthropic slightly than deal with any real safety concern.
Business organisations
This incident has equally sparked dialog inside industrial organisations that depend on US-owned AI fashions.
Mary Mesaglio, analyst at enterprise administration consultancy Gartner, stated this incident emphasised the “want for organisations to be intentional about sovereignty dependencies and, the place doable, to design model-agnostic architectures”.
She added that CIOs and AI leaders should stabilise AI-dependent workflows now and use this occasion to deal with mannequin focus threat, talent-technology resilience and sovereign AI disruption.
From a cyber safety perspective, Jamie Moles, a senior technical supervisor at US-based AI cyber safety firm ExtraHop, stated: “The priority over a ‘jailbreak’ that uncovers software program flaws highlights why a defence-in-depth technique is non-negotiable.”
He added: “Excellent mannequin resistance doesn’t exist, and it’s necessary to by no means depend on only one mannequin, as an motion like this will create important setbacks for the person consumer and wider tasks throughout an organisation.”
