AI firms amplify existential fears to shield operations from regulation, investigation finds
A BBC Future inquiry suggests major technology companies are leveraging fear of their own creations to discourage external oversight while highlighting immediate issues like environmental damage and labour exploitation.
An investigation by BBC Future suggests that leading artificial intelligence firms, specifically Anthropic and OpenAI, are strategically emphasising catastrophic risks regarding their own technologies to shape public perception. This approach involves promoting narratives of existential danger, such as the cybersecurity capabilities of Anthropic's new model, Claude Mythos, while simultaneously releasing or selling the technology to the market.
The report highlights that executives, including Dario Amodei of Anthropic and Sam Altman of OpenAI, have historically issued warnings about unreleased models before proceeding with their launch anyway. For instance, in 2019, Altman warned against releasing GPT-2 due to safety concerns, yet the model was released months later. Critics argue this pattern of fear-mongering serves to distract from current, tangible harms caused by AI deployment, such as environmental destruction, labour exploitation, and mental health impacts.
By framing AI development as an unmanageable danger, these companies aim to discourage public regulation and reinforce the notion that only they possess the expertise to ensure safety. The narrative of an inevitable apocalypse makes regulators appear powerless, encouraging a belief that the tech giants themselves are the only entities capable of preventing disaster. This strategy may also serve to stabilise stock prices as these institutions pursue public trading status.
Experts question the validity of claims regarding the power of new tools like Mythos, noting a distinct lack of industry-standard metrics to validate them. Heidy Khlaaf, chief AI scientist at the AI Now Institute, pointed out the absence of false positive rates, a standard measure for security tools. While Anthropic claimed Mythos found thousands of high-severity vulnerabilities, it did not measure the tool against existing benchmarks or provide data on its error rates, leading to significant scepticism regarding its actual capabilities.
The investigation notes a contradiction between the founding narratives of these firms and their current commercial ambitions. Anthropic was established as a splinter group from OpenAI, citing a lack of dedication to safety in the parent company, yet both are now pursuing public trading status. Furthermore, Anthropic has abandoned its previous policy of never training a model without guaranteed safety measures, and Google has reportedly dropped its red lines regarding the development of AI weapons.
Ultimately, the piece concludes that framing AI as an existential threat discourages meaningful governance. As one expert noted, if the technology is portrayed as capable of ending the world, immediate problems like greenhouse gas emissions from data centres or the psychological effects on vulnerable populations seem insignificant by comparison. The strategy ensures that the public feels unable to intervene, leaving the fate of the technology in the hands of the very corporations building it.
