Anthropic co-founder Jack Clark has said governments may eventually need to require artificial intelligence companies to maintain independently verifiable “kill switches” capable of shutting down dangerous AI systems.
Clark argued that as AI models become more powerful and potentially more autonomous, there should be a reliable way to stop systems if they begin behaving in ways that could pose serious risks.
The proposal would go beyond companies simply having internal emergency controls. Clark suggested that any shutdown mechanism should be verifiable by an independent third party, helping regulators establish whether companies could actually disable a system when necessary.
The comments come amid growing debate over the safety of increasingly capable AI models. Anthropic CEO Dario Amodei has also called for the industry to slow the development of frontier AI systems so that safety measures can keep pace with technological progress.
Supporters of stronger safeguards argue that AI systems could eventually operate with greater independence, making emergency controls an important part of risk management. However, experts also face technical challenges in designing a shutdown mechanism that cannot be bypassed by a sufficiently capable system.
The debate is also reaching governments. US lawmakers are considering proposals that would give authorities greater powers to test AI companies’ safety measures and intervene when advanced systems are considered dangerous.
The discussion comes as AI companies face pressure to balance rapid innovation with concerns about cybersecurity, misuse and the possibility of advanced systems becoming difficult to control.
For now, a mandatory AI “kill switch” remains a proposal rather than a universal requirement. Clark’s comments highlight the growing push for independent oversight and stronger safeguards as the technology continues to advance.