OpenAI and Anthropic have entered into a voluntary agreement with the British government to allow intelligence agencies to assess the safety of their AI models. Under this arrangement, the firms have pledged to deactivate their platforms if British secret services identify them as unsafe. This mechanism is being referred to by some observers as a "kill switch," intended to mitigate potential risks such as the misuse of AI in cyberattacks or the development of biological weapons.

The agreement marks a significant step in the ongoing dialogue between major technology developers and national security officials regarding the rapid advancement of generative AI. While the companies have agreed to this oversight, the specific criteria for what constitutes an "unsafe" platform remain broad. The move is part of a wider effort by the UK government to establish itself as a global leader in AI regulation and safety testing. By granting intelligence agencies access to evaluate these systems, the companies aim to demonstrate a commitment to responsible development, though the practical implementation and the threshold for intervention have not been fully detailed. The collaboration highlights the tension between the fast-paced innovation of the private sector and the government's need to ensure national security in an era of increasingly powerful automated tools.