US kill switch bills would make frontier model shutdowns a continuity problem for customers

Gary Barlet, public sector CTO at Illumio, described what a switch-off looks like from the customer side. "If a model suddenly becomes unavailable, those workflows could be disrupted overnight," he told ITPro. "Even intended as a safety measure, it could become a source of uncertainty, disruption, and economic harm if it's triggered too aggressively or without clear standards," he said ^[9].

The timetable is long. Eight days after the Senate block, Republican congressman Tom Kean Jr introduced his own AI Emergency Button Act, on September 24 ^[4][19]. "While AI can be a helpful tool, it is essential that humans remain in control of complex artificial intelligence systems," Kean said in a press release ^[5]. ITPro reports it could be months before President Trump signs a bill into law ^[7].

Charlotte Wilson, head of enterprise at Check Point, raised the practical objection to any switch. AI runs on phones, physical networks, data centre networks and laptops, she told ITPro. "You'd have to find all of it before you could shut it down," she said ^[8]. Her objection applies to operators as much as to lawmakers. The Kiteworks survey result came in 19 percentage points worse than the company's own forecast ^[16]. That leaves 21% of respondents with a tested way to stop an active agent in real time ^[17].

Traceability is the other half of the package. "Right now, AI agents are running loose in our networks, and nobody can see them or verify who built them, making it increasingly hard to stop them," Democratic congressman Josh Gottheimer said in a press release ^[3]. The Stop Rogue AI Act gives the standards job to NIST ^[2]. ITPro's advice to companies starts with an inventory of the AI models they use ^[14]. The reporting does not say whether NIST's standards would bind companies that deploy agents or only guide them.

The exposure is uneven. Darren Guccione, CEO of Keeper Security, told ITPro that large companies can invest across a wide range of models, while smaller ones tend to depend on the same frontier models ^[11]. Smaller firms are also less likely to fund their own AI safety mechanisms, according to ITPro ^[15]. Guccione said a kill switch would not make AI secure by itself, though it would set an emergency baseline that frontier developers must meet ^[10]. He put the rest on the customer: "Continuity planning needs to treat that as an operational imperative in its own right." ^[13]

For a smaller operator, the trade-off is concentration against cost. One frontier model is cheaper to run; a second model, or a manual fallback, is what keeps a workflow going if the first is throttled ^[11]. I think the decision this quarter is narrower than the bills suggest: whether to build an agent inventory and test a stop before NIST defines either. Building now risks rework if the standards settle on a different method. Waiting keeps a company among Kiteworks' 79% on the day a developer is ordered to suspend a model ^[12].