Could a sudden loss of access to cutting-edge AI systems be the trigger for the UK's technological downfall? The recent decision by US-based Anthropic to temporarily restrict access to its top-of-the-line models, Mythos and Fable, has sparked fears that UK businesses might become reliant on American tech giants. But in an unexpected move, cybersecurity firm Darktrace is backing Anthropic's pause, urging a cautious approach to the deployment of advanced AI.
Anthropic's AI models are designed for tasks like identifying software vulnerabilities, but US officials imposed restrictions due to national security concerns about potential misuse. Access was reinstated only after additional safeguards were put in place. Darktrace's Senior Vice President of AI and Security Strategy, Nicole Carignan, believes the pause allowed a critical assessment of whether existing 'guardrails' were sufficient for a technology with significant weaponisation potential.
The incident has intensified concerns about 'AI sovereignty' – the UK's dependence on international allies for critical AI tools. The government's Science, Innovation and Technology Committee has warned ministers against over-reliance on overseas providers, citing the Anthropic restrictions as a prime example of how such reliance can lead to supply chain vulnerabilities.
Carignan advises organisations to manage frontier AI like other critical suppliers – diversifying their model usage across multiple providers. Darktrace itself works with various AI firms, including OpenAI, Gemini, and Microsoft, demonstrating this approach. A shift towards 'sovereign AI' is also anticipated, incorporating open-source models and domestic data centres to maintain control over sensitive data.
Legal experts see the incident as a significant development in government regulation of frontier AI. Caroline Ramsay notes that the temporary restrictions represent a material shift in US export controls, now extending beyond hardware like semiconductor chips to AI models themselves. Organisations heavily dependent on advanced AI should factor potential service interruptions and export controls into their risk management strategies.