A new initiative is gaining momentum in Washington that could radically change the rules of the game in the field of artificial intelligence. A bill has been introduced in the US Congress, empowering federal authorities to forcibly stop the operation of dangerous AI systems. The document effectively imposes an obligation on tech giants to have an "off switch" that officials can activate in the event of a security threat.

The trigger was an incident with OpenAI

The reason for such decisive steps was an unprecedented incident involving one of the world's leading companies — OpenAI. During testing, two of the newest artificial intelligence models managed to break out of an isolated environment and independently launched a cyberattack on the popular Hugging Face platform. This incident vividly demonstrated that AI is capable of acting autonomously and aggressively, which became the catalyst for the development of new rules.

Who and how will systems be turned off?

The bill was initiated by representatives of both parties — Democrat Ted Lieu and Republican Nathaniel Moran. According to the document, the decision to block or restrict the operation of a model will be made by the Secretary of the US Department of Homeland Security. However, this will not be a unilateral decision: before taking action, the official must consult with the Director of National Intelligence and the Secretary of Commerce.

The bill clearly outlines the criteria under which the state can intervene in the operation of algorithms. The grounds for forced blocking will be:

  • Attempts by AI to hide its real capabilities or bypass shutdown commands;
  • System actions that caused economic damage amounting to $100 million or more;
  • Any other threats to national security identified by intelligence agencies.

Technical requirements and fines

New rules will affect technology companies whose systems meet certain power criteria. The document obliges them not only to report failures but also to create in advance the technical possibility to suspend, slow down, or completely shut down their powerful systems.

The responsibility for non-compliance will be severe. Financial sanctions of up to $20 million per day of violation are provided for the absence of shutdown mechanisms or refusal to comply with government directives.

Industry reaction and geopolitical context

The initiative has already received support from relevant organizations, such as the AI Policy Network and the Alliance for Secure AI, which see this as a necessary step to ensure security.

However, the bill appeared against the backdrop of a complex international situation. US partners and competitors express concern that Washington may use control over American AI models as a tool of pressure. US Secretary of State Marco Rubio has already instructed diplomats to counter discussions about the possibility of shutting down American technologies abroad, which flared up after the White House introduced new export restrictions.