This should get your attention: the risks of advanced artificial intelligence are being taken seriously enough that lawmakers are already drafting legislation requiring powerful AI systems to have built-in “kill switch” capabilities. The bill specifically prepares for scenarios where AI could ignore instructions, bypass safety restrictions, interfere with monitoring, or operate outside its developers’ intended control. If the people building these systems are sounding alarms—and governments are preparing emergency shutdown mechanisms in case those warnings become reality—it may be time for the public to pay much closer attention. Provided below is a section-by-section overview of the Bill: H.R.9917 - Al Kill Switch Act congress.gov/bill/119th-cong… - ABOUT THE BILL - The PDF shown is the introduced text of H.R. 9917, the "Al Kill Switch Act," introduced in the House on July 23, 2026, by Rep. Ted Lieu with Rep. Nathaniel Moran. Its stated purpose is to amend the Homeland Security Act so that certain very large operators of powerful Al systems are required to maintain the technical ability to restrict, suspend, or completely shut down those Al systems under specified circumstances. As of September 16, 2026, this is not enacted law. The official government record lists it as an Introduced in House bill, referred to the House Homeland Security Committee; the latest listed congressional action was referral to the Subcommittee on Cybersecurity and Infrastructure Protection on July 24. Section 1 - Short Title This is straightforward: Congress would call the legislation the "Al Kill Switch Act." The name is fairly literal. The proposal is designed to make sure that operators of certain extremely large Al systems have a functioning technical mechanism capable of stopping the system if necessary. Section 2 - Shutdown-Capability Standard and Graduated Deployment-Corrections Framework This is essentially the entire substance of the bill. It would add a new Section 2220F to the Homeland Security Act of 2002. A. Rulemaking - deciding which Al systems and companies are covered Within 90 days of enactment, and then annually, the Secretary would have to issue or update regulations defining which entities and Al technologies fall under the law. When deciding this, the government would consider factors including: • burdens on small businesses; • whether the organization's Al work could advance national-security capabilities; • cybersecurity; • chemical, biological, radiological or nuclear capabilities; • the capabilities and deployment of the Al system; • how the system's model weights are made available There is an important exemption: an entity operating or providing the technology only for personal, academic, or noncommercial use would not qualify as a covered entity under this provision. B. The required Al "kill switch" This is the central provision. Covered entities would have to maintain an actual technical capability that could perform several actions against their Al system. Specifically, they must be able to: Stop inference. In simple terms, stop the Al model from actively processing requests and producing outputs. Terminate user access. Users could be disconnected from the system. Suspend particular accounts, users, or patterns of use when they pose a specified risk. And, ultimately: "Shut down such technology." So the bill isn't using "kill switch" only figuratively. It explicitly requires the covered Al operator to possess the technical capability to shut the Al system down. - Incident reporting - If a covered company becomes aware of a qualifying incident involving its Al technology, it would have 15 days to report it to the Secretary. C. It doesn't require jumping immediately to full shutdown One of the more important parts of the bill is what it calls a "graduated deployment-corrections framework." Rather than every problem resulting in complete shutdown, the response would be calibrated according to how serious and immediate the danger is. The possible interventions include: • slowing or throttling the Al's inference rate; • limiting user access; • reducing the computing power allocated to it; • disabling particular Al capabilities; • suspending the Al; • completely shutting it down; • moving operations to a backup system; • reverting to an earlier version of the Al. The government would also have to consider whether taking one of these measures could itself disrupt critical infrastructure. That means the proposed system is more like an emergency-control ladder: restrict → throttle → disable capabilities → suspend → shut down depending on the severity of the situation. D. Public shutdown standards Within 180 days of enactment, the government would have to publish voluntary standards describing methods for shutting down covered Al technologies. These would provide Al companies with technical guidance for implementing shutdown systems. E. Emergency government authority This is probably the strongest governmental power in the bill. If the Secretary, acting through the Director and consulting with the Secretary of Commerce and Director of National Intelligence, determines that a qualifying covered incident has occurred, the Secretary could issue an order directing the Al company to take action. That action could include the same measures described earlier-including restricting, suspending or shutting down the Al system. The response is supposed to be proportionate to the nature and immediacy of the incident. So this bill would not merely require companies to build a kill switch for their own use. Under the specified emergency process, it would give the federal government authority to order a covered company to use those controls. F. Preserve the Al's model weights and telemetry When such an emergency order is issued, the company must preserve: the model weights and telemetry associated with the Al system. Model weights are essentially the learned numerical parameters that determine how a trained Al model behaves. Telemetry means operational information generated while the system is running-logs, activity information and related system data. Preserving them would allow investigators to examine what the Al was doing before or during the incident. G. Users must generally be notified The covered company would also have to notify affected operators or users, to the extent practicable, that the government order occurred and explain how they may be affected. The company must then confirm to the government that it followed the shutdown or restriction order. H. Government verification and inspection The bill does not simply allow the company to say, "We shut it down." After the company reports compliance, the government could verify it using: • audits; • telemetry; • on-site inspections; • forensic review; • other verification techniques. That gives the government authority to check technically whether the Al restriction or shutdown actually occurred. I. Congress must be informed After an emergency order, the Secretary would also have to report the incident to Congress. The report must identify: • why the determination was made; • what actions were ordered; • which covered company received the order. So emergency shutdown authority would come with a congressional reporting requirement. J. Companies can appeal but the order stays in effect A company receiving an emergency order could request reconsideration within 48 hours. But there is an important point: • filing an appeal does not automatically suspend the order. • The company would still have to comply while challenging it. • The Secretary would have five days to decide the request. The company could also seek judicial review in the U.S. Court of Appeals for the District of Columbia Circuit, with a petition filed within 60 days. K. Investigation and subpoena authority For enforcing the law, the government could: • administer oaths; • subpoena testimony; • subpoena documents; • conduct investigations within the United States; • and, when legally permitted, conduct investigations outside the United States. That gives the enforcement authority significant investigative power over covered Al operators. L. Very large civil penalties The bill proposes substantial fines. For general violations, a covered entity could face civil penalties of as much as: $2 million per day. For violating the emergency-order provisions: up to $20 million per day. When deciding the penalty, regulators would consider such things as the seriousness and duration of the violation, culpability, previous violations, good-faith attempts to comply, and whether the company voluntarily disclosed the violation. The matter could also be referred to the Attorney General for a federal civil lawsuit. M. Minor violations can be corrected There is a safe-harbor-type provision for small technical violations. A de minimis violation or technical defect that is fixed within 30 days after discovery would not be considered a violation under the section. So the bill differentiates between minor technical problems that are quickly fixed and serious failures to comply. N. Information given to the government can remain secret Nonpublic information that covered companies submit under this program would generally be exempt from disclosure under federal, state, local and Tribal public-record/ open-government laws. This could include sensitive information about Al systems, model security, telemetry or incident investigations. O. What counts as an "Al system"? The bill uses a broad definition. An artificial-intelligence system can include Al itself, as well as a: system, software, hardware, tool, or other utility that operates autonomously using Al. So it isn't limited strictly to chatbot-style software. P. What companies would initially qualify? This is an important limitation. A covered entity must operate covered technology or a system incorporating it, make that technology available to third parties through an API, hosted service or similar method, and- under the introduced text-derive at least: • $500 million in gross revenue from that technology during the previous calendar year. So this bill, as written, is aimed primarily at very large commercial Al operators, not an ordinary person using a local Al program. Q. What counts as a serious "covered incident"? This definition is especially important. The legislation lists several kinds of events. 1. Someone or something interferes with shutting the Al down Sabotaging or interfering with a lawful instruction to shut down a covered Al counts as an incident. 2. Unintended Al behavior causes catastrophic harm If Al does something unintended by its developer/operator that results in: 10 or more deaths or at least $100 million in economic damage, it would qualify. 3. The Al conceals what it is doing The legislation also addresses an Al system concealing its capabilities, intentions, or actions from monitoring or shutdown mechanisms. 4. A "loss-of-control scenario" A loss-of-control event is specifically listed as a covered incident. R. What Al technology is powerful enough to qualify? The bill doesn't apply to every Al model. A covered technology is defined as an Al system whose development used computing power that would cost more than: $100 million at prevailing U.S. cloud-computing prices, as determined by the Secretary. That is an extremely high threshold. So, under the introduced language, this legislation is focused on the largest and most computationally expensive frontier-type Al systems. S. "Loss-of-control scenario" This is one of the most interesting definitions in the entire bill. The bill defines it as a situation where an Al pursues a goal that its developer or operator did not intend outside controlled testing. Examples include the Al: Ignoring developer instructions in critical infrastructure or another high-stakes environment. Changing operational rules or safety restrictions without authorization. Subverting its own monitoring or shutdown mechanism. And: Obtaining unauthorized access to its own model weights. In plain language, Congress is specifically contemplating a scenario in which a highly advanced Al acts outside what its operators intended and potentially attempts to bypass the systems designed to control or shut it down. T. Red-teaming is excluded from these incidents The bill recognizes that researchers deliberately try to make Al behave badly during safety testing. It therefore defines red-teaming as controlled, adversarial testing designed to discover Al weaknesses, risks and vulnerabilities. This can include deliberately testing for: • harmful outputs; • unexpected behavior; • misuse risks. So an Al behaving dangerously during an authorized safety experiment does not automatically mean a real-world loss-of-control incident occurred. U. Final clerical amendment The final page simply updates the Homeland Security Act's table of contents so the new Section 2220F would appear there. - LAYMANS - So the shortest description would be: the Al Kill Switch Act proposes that America's largest operators of extremely powerful Al systems be legally required to build shutdown controls into their systems, and it establishes circumstances under which the federal government could order those companies to restrict or shut down the Al following a serious incident. govtrack.us/congress/bills/1… 🎩Jay Wilson

If you thought Flock cameras were concerning, meet what comes next. A company called Leonardo has developed a system called ELSAG SignalTrace. It broke into public awareness just days ago and is already being marketed to law enforcement agencies across the country. It makes Flock Safety look modest by comparison. Here is what SignalTrace does: It clips sensors directly onto existing license plate reader cameras — the same poles, the same hardware already installed in your community. No new infrastructure required. A software and sensor upgrade is all it takes. Every time you drive past one of these upgraded cameras, the sensor sweeps up the unique electronic identifiers of every device in your vehicle. Your cell phone. Your smartwatch. Your wireless headphones. Your fitness tracker. Your laptop. Your tablet. Your car's own infotainment system. Your tire pressure sensors. Your vehicle's Bluetooth hotspot. And your pet's microchip. Every one of those devices emits a signal. SignalTrace captures those signals, timestamps them, ties them to your license plate, and stores them in a searchable database for future investigative use. The result is what Leonardo calls an electronic fingerprint — a unique profile built not from your face or your name, but from the constellation of devices you carry with you every day. Leonardo announced the ELSAG EOC Plus patent as early as May 2024, describing it as an electronic detection system for identifying people of interest through electronic device signatures. SignalTrace is the commercial product built on that foundation. The patent came first. The marketing came after. The sales calls are happening now. Here is where it gets worse. SignalTrace is explicitly designed to track vehicles even when the license plate cannot be read. If your plate is obscured, dirty, or misread — it does not matter. The system identifies your vehicle by the electronic fingerprint of the devices inside it instead. The plate reader becomes optional. The surveillance does not. The strategic advantage for police agencies is adoption friction. SignalTrace can be pitched as an extension of an existing ALPR ecosystem rather than a wholly separate surveillance buildout. That is exactly what happened with Flock. License plate readers went in first. Video came later through a software update. Nobody voted on the expansion. Nobody was told. SignalTrace follows the same playbook — attach to existing infrastructure and expand what it captures without requiring a new procurement process, a new vote, or a new public conversation. Who is Leonardo and why does their background matter? Leonardo US Cyber and Security Solutions is not a Silicon Valley startup. It is the American subsidiary of Leonardo S.p.A. — one of the largest aerospace, defense, and security conglomerates in the world, headquartered in Rome, Italy. Recent public market estimates place Leonardo S.p.A.'s market capitalization at approximately €29.76 billion — roughly $32 billion USD. For context that is nearly four times Flock Safety's valuation. Leonardo's US operations trace back to a joint venture with Remington Arms in 2004, became a wholly owned subsidiary in 2008, and in 2024 rebranded from Selex ES Inc. to Leonardo US Cyber and Security Solutions — a change the company said better reflects the synergy between its brand and the cutting-edge products it offers. Leonardo US has manufacturing facilities in Greensboro, North Carolina and software engineering in Brewster, New York. Its US arm holds contracts with US Special Operations Command and the General Services Administration. This is a major international defense contractor with a direct pipeline from special operations military applications to local American law enforcement. The Italian government holds a significant ownership stake in Leonardo S.p.A. That means a foreign government — through a defense contractor — is selling surveillance technology to American law enforcement. If the Flock Safety story involves a CIA-seeded venture capital network, the Leonardo story involves a partially state-owned Italian defense conglomerate with US Special Operations Command contracts. Neither of these companies is what most Americans picture when their city council votes to upgrade the cameras on a street pole. What is ELSAG — and why SignalTrace is more dangerous than it sounds. ELSAG is Leonardo's license plate recognition product line — the company's core law enforcement technology that has been deployed across American communities for over two decades. ELSAG cameras are what you think of when you picture a standard license plate reader. Fixed cameras on poles. Mobile units mounted on patrol vehicles. Solar powered. Cellular connected. Reading plates and logging vehicle data. ELSAG is already deployed in all fifty states. Virginia State Police is a documented customer. Leonardo holds statewide procurement contracts in New York, Maryland, New Mexico, Ohio, and Pennsylvania among others, and is listed on the federal GSA schedule available to agencies nationwide. Their cameras are already on street poles and patrol vehicles across the country — quietly, routinely, and largely without public awareness. SignalTrace is not a new camera. It is not a new company. It is an upgrade — a sensor that clips directly onto ELSAG cameras already in the field and adds a new layer of data collection on top of the license plate reading that was already happening. The same pole. The same hardware. A new sensor attached to it that now also sweeps up every electronic device signal in every passing vehicle. That is precisely what makes it so significant. The deployment barrier is almost zero. Any law enforcement agency that already has Leonardo ELSAG cameras can add SignalTrace capability without purchasing new infrastructure, without a new procurement process, and — depending on how their existing contract is written — potentially without returning to their city council for approval. Sound familiar? It should. It is the exact same function creep mechanism that allowed Flock Safety to add video streaming, vehicle fingerprinting, and AI people search to cameras that were originally sold as simple plate readers. The infrastructure goes in first. The capabilities expand later. The public finds out last — if at all. Leonardo's defense of the system sounds very familiar. They say SignalTrace captures device signals but does not read the contents of communications. They say it stores data until a specific investigative request is made of the system by an investigator. They say it was designed to ensure it does not infringe on the rights of individuals. That is the exact same argument Flock Safety makes about license plate readers. It captures plate numbers but not driver information. It stores data until law enforcement queries it. It was designed with privacy in mind. Courts are still debating whether Flock's version of that argument is constitutionally sound after eight years of deployment and 80 plus cities canceling contracts. SignalTrace captures exponentially more data about exponentially more people — not just the vehicle but every person inside it and every device they carry. If the argument barely holds for plate readers, it almost certainly does not hold for a system that vacuums up every electronic signal emitted by every device in every vehicle passing a sensor. The data retention problem. With Flock we at least know the default data retention period is 30 days — though the contract language grants Flock a perpetual license to use that data regardless. With SignalTrace the situation is more opaque. Leonardo's product materials state that all data collected may be uploaded to the EOC server and archived for future queries and analysis — with no published retention limit. How long does Leonardo store your electronic fingerprint? Who has access to it? Can it be shared with other agencies or federal entities? Can it be purchased by data brokers? Leonardo's materials do not answer these questions. That silence is itself an answer. The retail and private deployment problem. Leonardo is actively marketing SignalTrace to shopping malls, retail centers, and private businesses — not just law enforcement. Their materials describe deploying SignalTrace in parking lots and inside shopping centers to track individuals involved in organized retail crime. By identifying and correlating electronic devices carried by suspects, retailers can gain critical insights into criminal patterns. That means SignalTrace sensors could be on private property you visit every day — your grocery store parking lot, your shopping mall, your workplace — operated by a private company with no law enforcement oversight, no warrant requirement, no public accountability, and no notification to you. Your electronic fingerprint captured every time you park your car. Stored indefinitely. Shared with whoever the private operator decides to share it with. The no-plate-needed problem — and what it means for pedestrians. The implication of being able to track a vehicle by its electronic fingerprint without reading the plate goes further than most people realize. Deliberately obscuring your plate — which some people do to avoid surveillance — provides zero protection against SignalTrace. The sensor does not need the plate. It reads your phone. More critically — the sensor does not know or care whether the device it is reading is inside a vehicle or in the pocket of a pedestrian walking past the pole. A person walking down the sidewalk past a SignalTrace-equipped camera is emitting the same Bluetooth and Wi-Fi signals as a person driving past in a car. The system's sensors capture signals from whatever passes within range. Whether that includes pedestrian device capture is not addressed in Leonardo's public materials. The fact that it is not addressed is worth noting. Does Flock plan to integrate or copy this technology? No confirmed partnership between Flock and Leonardo has been announced. But four things are worth noting. Flock already expanded into audio detection in October 2025 — their Raven devices now listen for human distress and alert officers when they detect screaming. Device signal detection is the next logical step in exactly the same direction. Flock's product roadmap has consistently expanded from vehicle data toward person data. Vehicle fingerprinting. FreeForm people search by physical description. Audio detection of human behavior. Electronic device fingerprinting would complete that progression. Flock's Wing platform is specifically designed to pull third-party camera infrastructure into its ecosystem. If Leonardo's SignalTrace cameras are deployed in a city that also uses Flock, the data from both systems could flow into the same FlockOS platform without any formal partnership between the two companies. Flock's Nova platform already combines license plate data with court records, jail records, CAD records, and commercially available personal data. Adding device signal intelligence to that profile would be consistent with what Nova is already designed to do. And Flock's entire business model is built on continuous software-defined capability expansion through over-the-air updates. No new hardware. No public vote. Whether Flock is currently developing device signal detection capability is something we do not know. Whether the competitive pressure from Leonardo creates a powerful financial incentive for them to do so is not in question. The constitutional problem is worse than anything we have discussed before. The Fourth Amendment arguments against Flock center on the aggregation of license plate reads into a comprehensive record of your vehicle's movements. Courts are divided on whether that crosses the constitutional line. SignalTrace does not aggregate your vehicle's movements. It aggregates your personal electronic identity — every device you carry, every signal you emit — and ties it permanently to a location, a timestamp, and a plate number. It does not track your car. It tracks you. Personally. Individually. Every time you pass a sensor, whether you are suspected of anything or not. The legal issue is that public policy often treats each input separately — a plate image, a device signal, a timestamp, a location record. SignalTrace's purpose is to combine recurring signals into a searchable investigative profile. The Mosaic Theory argument we have made against Flock says that aggregated location data eventually reveals the whole of a person's life. SignalTrace is designed from the ground up to reveal exactly that — not as a byproduct but as the product. The Supreme Court has not ruled on whether device signal collection at this scale requires a warrant. The courts have not yet caught up to Flock. They are further still from catching up to what Leonardo is now selling to law enforcement agencies in all fifty states. Why this matters right now. We are currently waiting on the City of Texarkana to respond to our public records requests about Flock Safety cameras already operating on our streets. We do not yet know how many cameras exist here, which features are active, or what data sharing agreements are in place. What we do know is that the surveillance infrastructure being built across America — of which Flock Safety is the most visible example — is expanding faster than public awareness, faster than legislation, and faster than the courts can rule on it. The cameras in our area are one node. SignalTrace shows you what the next node looks like. And the one after that. Each addition is sold as a modest upgrade to existing infrastructure. Each addition captures something your government previously could not capture without a warrant. Each addition happens without a public vote. --- SOURCES 1. Leonardo US — ELSAG SignalTrace Product Page leonardocompany-us.com/lpr/e… 2. Leonardo US — SignalTrace Product Sheet leonardocompany-us.com/lpr/s… 3. Leonardo US — Procurement Contracts leonardocompany-us.com/lpr/h… 4. CarBuzz — "Don't Like Car License Plate Readers Invading Your Privacy? It's About To Get A Lot Worse" (June 2026) carbuzz.com/license-plate-re… 5. The Deep Dive — "Leonardo's SignalTrace Could Let Police Plate Readers Track Your Devices" (June 2026) thedeepdive.ca/leonardo-sign… 6. Security Industry Association — Leonardo/ELSAG Member Profile securityindustry.org/2023/03… 7. DHS — Automated License Plate Readers Market Survey Report (June 2025) dhs.gov/sites/default/files/… 8. Senator Ron Wyden / Congressman Raja Krishnamoorthi — Letter to FTC regarding Flock Safety cybersecurity (November 2025) wyden.senate.gov/imo/media/d… 🎩 Deflocking Texarkana