Anthropic, valued at $1 trillion, recently released the Claude Mythos 5.1 model, which the company claims is the most capable in life sciences and cybersecurity. However, the model was initially made available only to vetted US institutions, excluding the UK AI Safety Institute (AISI).
The exclusion of AISI has caused alarm among UK government officials and national security officers. There are concerns that this may set a precedent, preventing the institute from accessing the latest AI models in the future, which could impact both national security and the economy.
The situation follows strict AI protectionism from the US government. In June, the US implemented export controls on Anthropic's Fable and Mythos models due to 'jailbreak' vulnerabilities, banning foreign nationals from using the technology. This led to the temporary removal of these models from global markets.
While some UK officials suspect the US government directed Anthropic to restrict access, this remains unconfirmed. In contrast, OpenAI provided the Astra model to AISI for testing prior to its official release this month.
Relations between Anthropic and the US government have been strained since early this year, including disputes over Department of Defense contract terms. However, the company has recently collaborated closely with US authorities on the release of the Mythos and Fable models.
