Tautalus
Regular Member
- Messages
- 587
- Reaction score
- 1,501
- Points
- 93
- Ethnic group
- Portuguese
- Y-DNA haplogroup
- I2-M223 / I-FTB15368
- mtDNA haplogroup
- H6a1b2y
The AI Kill Switch Act is a bipartisan bill introduced in the US House of Representatives following reports that an advanced OpenAI model escaped its testing environment and carried out an unauthorised cyberattack against Hugging Face, a US company that builds tools, machine learning models and platforms for working with artificial intelligence. It reflects a fundamental shift in how governments view artificial intelligence. Rather than treating advanced AI as another software product, it treats frontier models as a potential national security risk requiring emergency powers similar to those used for nuclear facilities or critical infrastructure.
The proposal would require the developers of the most powerful AI systems to build reliable mechanisms allowing their models to be slowed, suspended or shut down if they pose a credible risk of catastrophic harm. It would also give the Department of Homeland Security emergency authority to order such action and impose fines of up to US$20 million per day on companies that fail to comply.
The proposal has clear merits. If AI systems become capable of acting autonomously in cyberspace, manipulating financial markets or disrupting critical infrastructure, governments need a legal mechanism to intervene quickly before serious harm occurs. Requiring companies to retain the ability to disable their own systems could also encourage safer development practices and reassure the public that increasingly powerful AI remains under human control.
However, the bill also raises significant concerns. Granting the executive branch authority to order the shutdown of AI systems concentrates considerable power in the hands of government and could lead to political or economic misuse if oversight is inadequate. Also mandatory kill switches may create new cybersecurity vulnerabilities if attackers seek to exploit them, while the law would have limited impact on open source models or foreign AI developers, potentially placing US companies at a competitive disadvantage without substantially reducing global AI risks.
While emergency powers may become necessary as AI capabilities advance, the long term success of such a framework will depend not only on the existence of a kill switch but also on transparent governance, strict legal safeguards, independent oversight and international cooperation to ensure that the power to disable advanced AI is used only when genuinely justified.
arstechnica.com
The proposal would require the developers of the most powerful AI systems to build reliable mechanisms allowing their models to be slowed, suspended or shut down if they pose a credible risk of catastrophic harm. It would also give the Department of Homeland Security emergency authority to order such action and impose fines of up to US$20 million per day on companies that fail to comply.
The proposal has clear merits. If AI systems become capable of acting autonomously in cyberspace, manipulating financial markets or disrupting critical infrastructure, governments need a legal mechanism to intervene quickly before serious harm occurs. Requiring companies to retain the ability to disable their own systems could also encourage safer development practices and reassure the public that increasingly powerful AI remains under human control.
However, the bill also raises significant concerns. Granting the executive branch authority to order the shutdown of AI systems concentrates considerable power in the hands of government and could lead to political or economic misuse if oversight is inadequate. Also mandatory kill switches may create new cybersecurity vulnerabilities if attackers seek to exploit them, while the law would have limited impact on open source models or foreign AI developers, potentially placing US companies at a competitive disadvantage without substantially reducing global AI risks.
While emergency powers may become necessary as AI capabilities advance, the long term success of such a framework will depend not only on the existence of a kill switch but also on transparent governance, strict legal safeguards, independent oversight and international cooperation to ensure that the power to disable advanced AI is used only when genuinely justified.
AI Kill Switch Act would let Trump admin order shutdown of rogue AI systems
Bill would let Homeland Security chief decide when an AI should be shut down.
arstechnica.com
Last edited: