# Bill Gates says an AI "kill switch" alone will not make AI safe > Bill Gates told NBC a legally required kill switch will not make AI safe on its own, days after the Senate nearly passed a bill requiring one. FireAI Security & Research Team (HisnLabs) · Published 2026-09-28 Canonical: https://hisnlabs.com/en/news/bill-gates-ai-kill-switch Bill Gates said a legally required "kill switch" will not, by itself, make artificial intelligence safe, in an interview broadcast on NBC's Meet the Press on 27 September 2026 [[2]](https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979). The remark came days after the US Senate nearly passed a law requiring AI companies to build one into their most powerful systems [[1]](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236). ## Background Ten days before Gates's interview, Sen. John Neely Kennedy (R-La.) tried to pass a bill requiring AI companies to build an emergency shutdown mechanism into their most powerful systems, giving human operators inside those companies authority to disable a model judged dangerous. Sen. Rand Paul (R-Ky.) blocked the request on the Senate floor, arguing that regulating a technology lawmakers do not yet understand "risks killing innovation and setting it back decades," and proposed a study committee instead; Kennedy rejected that amendment and the bill did not advance [[3]](https://hoodline.com/2026/09/rand-paul-blocks-ai-kill-switch-bill-warns-of-innovation-wipeout/). A separate House bill, H.R. 9917 from Reps. Ted Lieu and Nathaniel Moran, would let the Department of Homeland Security order a slowdown or shutdown after a serious AI-linked incident, with daily fines for non-compliant companies [[1]](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236). ## Findings > The kill switch is, that’s kind of a weird thing. It’s not enough to have a kill switch. > > — Bill Gates, on NBC’s Meet the Press Gates told host Kristen Welker that today's AI models "are not yet at the point where they autonomously grab computers and... can't be shut down," and said the debate over an off-switch "shows how nontechnical various people are" about the actual risk [[2]](https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979). The danger he described is people misusing AI for cyberattacks or bioterrorism, not a model acting on its own, which an on-off button does not address [[2]](https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979). His alternative is legally mandated monitoring and record-keeping inside AI companies, so regulators can see what a powerful model is being used for. "No one thinks self-regulation is enough," he said, adding that the United States should set the standard because other countries, including China, will likely follow [[1]](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236). ## Implications for Mac users This is an argument about what the largest AI labs must build into their most powerful models, and how Washington regulates them, not about anything running on a personal computer. Nothing in this debate changes what an app on a Mac can do today. It is a preview of what "AI safety" is likely to mean in law when it arrives: mandatory monitoring and logging at the company level, decided by lawmakers and regulators, not a switch an individual user would ever touch [[1]](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236) [[2]](https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979). > The kill switch Congress is arguing about lives inside AI labs, not on a desk. Test FireAI free for 17 days to see, and cut off, exactly what a Mac connects to. [Download FireAI for Mac](https://hisnlabs.com/en/download) ## Recommendations 1. Treat this as background rather than a checklist: no Senate bill, passed or failed, changes anything that needs configuring on a personal device. 2. For anyone using AI tools day to day, ask what each one logs and where, since company-level monitoring is what Gates is asking Congress to require of the labs themselves. 3. Notice which apps on a Mac already act like small AI agents, booking things, filling in forms, reading files, and grant them only the access they need. 4. Anyone following where this goes should watch H.R. 9917 and any Senate follow-up rather than this single interview, since that is where an actual rule would come from. ## Relevance to FireAI The phrase "kill switch" is worth being precise about, because FireAI has one too, and it is not the mechanism Gates and Congress are discussing. The kill switch in the Senate debate is built into an AI lab’s own infrastructure: a way to halt a specific frontier model at the company level, potentially on a regulator’s order. It is a governance and engineering question for organisations that train and run large AI models, not a feature on a personal computer. FireAI's [kill switch](https://hisnlabs.com/en/docs/kill-switch) is a smaller, more literal thing: a button in the menu bar that cuts one Mac off from the internet. Turning it on refuses new connections to the outside world; devices already talking to each other on the same network, such as a printer or a shared drive, keep working. It does not close connections already open, and it stops working while Protection is paused. It does not watch, log or halt what any AI model is doing anywhere, on the Mac or in a lab's data centre, and has no connection to the shutdown authority Kennedy's bill or H.R. 9917 would hand to companies or regulators. The closest thing FireAI offers to an everyday version of "monitoring" is [security modes](https://hisnlabs.com/en/docs/security-modes) and [connection prompts](https://hisnlabs.com/en/docs/answer-your-first-connection-prompt): they show which apps on a Mac connect where, and let the person using it decide, rather than a company or a government deciding for them. That is a smaller, more local job than the one Gates describes, and it is not an AI safety mechanism in his sense. > Gates wants AI labs to log what their models are doing. FireAI already shows, app by app, what a Mac is doing on the network, one click from a full stop. [Download FireAI for Mac](https://hisnlabs.com/en/download) ## Limitations None of the sources says what specific monitoring and record-keeping requirements Gates has in mind beyond the general idea, or how such a rule would be enforced across companies outside US jurisdiction. Politico and Hoodline describe the Senate fight from different angles and do not fully overlap on procedural detail, and neither source confirms whether H.R. 9917 has scheduled hearings or a realistic path to a vote [[1]](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236) [[3]](https://hoodline.com/2026/09/rand-paul-blocks-ai-kill-switch-bill-warns-of-innovation-wipeout/). Try [FireAI, by HisnLabs](https://hisnlabs.com/en/download) free for 17 days. ## Sources - [Politico (Jalen Beckford), 27 September 2026: Bill Gates says an AI "kill switch" isn’t enough](https://www.politico.com/news/2026/09/27/bill-gates-ai-kill-switch-01094236) - [NBC News, Meet the Press, 27 September 2026: Bill Gates says it’s "not enough to have a kill switch" for AI (full interview)](https://www.nbcnews.com/meet-the-press/video/bill-gates-says-it-s-not-enough-to-have-a-kill-switch-for-ai-full-interview-270615621979) - [Hoodline, 17 September 2026: Rand Paul blocks AI kill switch bill, warns of innovation wipeout](https://hoodline.com/2026/09/rand-paul-blocks-ai-kill-switch-bill-warns-of-innovation-wipeout/)