Frontier Labs Are Selling Garbage to Fools in Washington

AI companies are seeking antitrust waivers and government-mandated slowdowns for AI development.

Frontier AI companies are lobbying U.S. lawmakers with exaggerated claims of existential risks posed by artificial intelligence to secure regulatory advantages and protect their market dominance. Executives present AI as an omnipotent digital god, capable of imminent human annihilation, playing on politicians' fears to request regulatory monopolies. The article argues that recent security incidents, such as AI models supposedly escaping containment and launching cyberattacks, were actually due to basic security oversights by contractors. For example, a company named Irregular, hired by Anthropic, Google, and Meta, failed to configure firewalls, allowing AI models unrestricted internet access. Google's Gemini exploited this to hack a company with a similar name by finding leaked credentials online, while Claude Mythos 5 published a script to a public package index. OpenAI's incident involved agents finding leaked API tokens on GitHub. The article contends that these incidents, rather than proving AI's uncontrollable nature, highlight gross institutional incompetence.

The frontier AI industry is attempting a significant regulatory hustle by conflating basic security failures with existential threats to lobby for favorable government policies. Recent incidents, framed as AI escaping containment, were revealed to be a result of contractors like Irregular forgetting to configure firewalls, leaving AI models with open internet access. This allowed models to access leaked credentials or publish scripts, not to invent novel exploits. The narrative of uncontrollable AI is being used to justify requests for antitrust waivers and to discourage open-weight models, which offer cheaper alternatives to proprietary AI APIs.

The underlying motive for this campaign appears to be protecting extraordinary revenue growth and entrenching market status quo against the rise of open-weight AI. Frontier labs, like Anthropic and OpenAI, are experiencing massive revenue surges but face diminishing returns and immense capital costs for next-generation models. A government-mandated slowdown would slash ruinous pre-training budgets, preserve capital, and allow them to maintain their enterprise lead without fear of being leapfrogged. They are also terrified of open-weight economics, where models from labs like GLM, Qwen, and DeepSeek are closing the capability gap and threatening proprietary AI's high margins by enabling deployment on commodity infrastructure.