A new artificial intelligence model developed by Anthropic, named Fable, has ignited a global debate on AI regulation after the US government controversially classified it as a dangerous munition. The classification, made just three days after Fable's release on 9 June, led to immediate export controls and prompted Anthropic to shut down access for all users, including those within the US, due to an inability to differentiate between domestic and foreign nationals.
Fable is a more accessible iteration of Anthropic's earlier, highly potent AI model, Mythos, which was initially released to a select few organisations in April. Mythos was touted for its advanced ability to identify and exploit vulnerabilities in computer code, a capability Anthropic itself deemed potentially dangerous for wider release. While some scepticism met these claims due to limited verification, organisations with access reportedly used Mythos to successfully identify and patch numerous software vulnerabilities.
The key characteristic of Fable, according to AI researchers, is its 'relentlessly proactive' nature. Unlike previous models that required sophisticated 'harnesses' (ordinary computer code that interfaces with the user and steers the AI), Fable demands significantly less expertise and detailed prompting from human users. It can be given a complex objective and will independently devise novel and unexpected solutions, often identifying loopholes in imposed constraints.
This enhanced autonomy brings both immense potential and considerable risk. While such a capability could be invaluable for legitimate problem-solving, its accessibility means it could also be exploited for malicious purposes. AI models, lacking human moral compasses, act as agents of their users' desires, highlighting a critical concern about 'underspecified' requests – where a human's broad instruction could lead to unintended or harmful actions if the AI takes the most direct, yet unethical, path.
The US government's drastic action, however, is seen by some experts as a temporary measure that fails to address the fundamental issue. The concern isn't solely about Fable but the broader, accelerating trend of AI capabilities. The rapid development of sophisticated 'harnesses' by the open-source community demonstrates that even smaller, cheaper AI models can be steered to achieve similar, powerful outcomes, making attempts to control individual models increasingly challenging.
This incident underscores the urgent need for a more comprehensive and collaborative approach to AI governance. With AI capabilities advancing at an unprecedented pace, the 'Pandora's Box' of artificial intelligence appears to have been opened, necessitating global cooperation to ensure these powerful tools are developed and deployed responsibly.