UK authorities are actively negotiating anthropic To enable limited deployment of the company’s advanced features Claude Mythos A.I. A platform for use by British companies. This effort reflects growing official awareness of the transformative but risky nature of cutting-edge AI in protecting digital systems from emerging developments. threats.
During the talks, British companies, especially financial services The industry oversaw early access to the tool, making the UK the only country outside the US to secure such regulations through the AI Security Institute.
anthropic Claude introduced the Mythos Preview in early April 2026 as its most advanced model to date.
Developed with a strong emphasis on cybersecurity practices, it quickly detects weaknesses in software and networks and completes complex tasks that often require significant time and effort from human experts.
Because of its potential, the company opted for a broad public disclosure.
Instead, it has launched controlled tests with a select group of trusted U.S. partners, including major tech firms and banks, as part of its Project Glasswing initiative, which focuses on defensive patching of vulnerabilities.
In Britain, preparations are underway to expand similarly vetted access to leading organisations. banks and other important events in the coming weeks.
This will allow testing in secure, isolated environments to assess the model’s ability to strengthen defenses without introducing new risks.
The initiative is based on the following foundations: EnglandProactive evaluation effort in which the Artificial Intelligence Security Institute (AISI) is currently conducting in-depth evaluations of Mythos Preview’s offensive and defensive potential.
These steps come with sharp warnings from senior figures. UK Technology Secretary Liz Kendall and Minister of Security And Jarvis recently sent an open letter to business leaders highlighting how Mythos and similar systems are accelerating cyber capabilities faster than expected.
They drew attention to that border artificial intelligence The relevant skills of models appear to double approximately every four months, encouraging organizations across all industries to prioritize basic hygiene measures such as regular updates, strong access controls and certification programs such as Cyber Essentials.
AISI’s testing confirmed that Mythos Preview is a significant advance, becoming the first model to fully navigate the demanding 32-step enterprise network attack simulation under controlled conditions.
While reviewers shined against vulnerable installations, they highlighted the limitations for well-protected environments and the dual-use nature of such technology; Valuable to defenders but potentially exploitable by adversaries.
Similar warnings have emerged internationally. WE As financial professionals were briefed on the model’s implications, European voices, including Germany’s central bank governor, described it as a “double-edged sword”.
Reports also show that: anthropic examines allegations of unauthorized preview access through a third-party channel and further highlights the need for tight controls.
For England Mythos can offer faster vulnerability detection and automated defense, especially for businesses that use sensitive data or critical infrastructure.
But significant progress depends on strict oversight, collaboration between government, industry and developers, and continued investment in core security practices. This is measured get engaged It demonstrates the UK’s strategy to embrace AI developments responsibly while effectively tackling the challenges head-on. digital transformation.
