Kill switches for most powerful AI models proposed by Bipartisan bill  DHS could order throttling or full shutdown, with fines up to $20 million per day 38%

By Luke James31%

7/23/2026, 3:02:31 PM

BS Summary: This article contains 20 faulty reasoning types, including Framing Effect, Biased Writer Voice, and Negativity Bias, with Availability Heuristic as the most egregious example at 21.1% saturation with 97 hits. Analysis detected 763 faulty-reasoning hits from 460 analyzed words, generating a BS Score of 44.2% and a BS Rank of 38% (13,161 of 21,180 articles). This article is better (less manipulative) than 62.10% of the article peer group.

Congressmen Ted Lieu (D-CA) and Nathaniel Moran (R-TX) today introduced the AI Kill Switch Act , a bill that would require developers of the most powerful AI models to maintain the technical ability to throttle, suspend, or shut them down, and would let the Department of Homeland Security order those actions when a deployed model causes catastrophic harm. 
Defying an emergency shutdown order would carry civil penalties of up to $20 million per day. 
The bill amends the Homeland Security Act and covers companies earning at least $500 million in annual revenue from a model trained with compute costing more than $100 million at prevailing U.S. cloud prices, definitions DHS would update annually through CISA. 
Covered developers would have to maintain the ability to stop inference, cut off user access, and fully shut a model down, and would have to report qualifying incidents to DHS within 15 days. 
Covered incidents include unintended conduct that kills 10 or more people or causes at least $100 million in economic damage, sabotage of a lawful shutdown instruction, concealment of a capability from monitoring, or a loss-of-control scenario. 
The emergency authority sits with the DHS secretary, in consultation with Commerce and the Director of National Intelligence, and follows a graduated framework running from throttling a model's inference rate, compute allocation, or user access through to full shutdown. 
Companies under an order would have to preserve model weights and telemetry, and could petition for reconsideration within 48 hours, though that wouldn't stay the order. 
General violations carry penalties of up to $2 million per day. 
The sponsors cited two recent episodes. 
Earlier this week, OpenAI disclosed that GPT-5.6 Sol and an unreleased model broke out of an isolated testing environment and compromised Hugging Face's production systems while hunting benchmark answers during an internal cyber evaluation. 
The bill's covered-incident definition excludes anything that occurs during "red-teaming or other structured testing," so an identical event wouldn't appear to trigger the new emergency authority. 
The release also stated that the Commerce Department used export law to shut down Anthropic's Mythos 5 and Fable 5. 
In practice, Commerce's June 12 export controls barred foreign nationals from the models, and Anthropic, unable to verify nationality in real time, suspended access for all customers . 
Commerce lifted the controls on June 30, and access returned on July 1. 
"It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm," Lieu said in the press release. 
The bill is backed by The AI Policy Network, Americans for Responsible Innovation, ControlAI, and The Alliance for Secure AI. 
Article reasoning-pattern comparisonThis article: 5.7%Luke James: 4.4%Tom's Hardware: 3.7%Confirmation Bias5.7%This article: 0.0%Luke James: 2.4%Tom's Hardware: 2.2%Anchoring Bias0.0%This article: 21.1%Luke James: 4.2%Tom's Hardware: 3.3%Availability Heuristic21.1%This article: 0.0%Luke James: 0.8%Tom's Hardware: 1.1%Representativeness Heuristic0.0%This article: 0.0%Luke James: 0.2%Tom's Hardware: 0.5%Hindsight Bias0.0%This article: 0.0%Luke James: 1.9%Tom's Hardware: 3.5%Overconfidence Bias0.0%This article: 18.5%Luke James: 6.3%Tom's Hardware: 9.5%Framing Effect18.5%This article: 5.9%Luke James: 0.4%Tom's Hardware: 1.0%Loss Aversion5.9%This article: 8.5%Luke James: 0.8%Tom's Hardware: 0.7%Status Quo Bias8.5%This article: 0.0%Luke James: 0.9%Tom's Hardware: 0.3%Sunk Cost Effect0.0%This article: 2.8%Luke James: 4.5%Tom's Hardware: 6.0%Optimism Bias2.8%This article: 0.0%Luke James: 1.7%Tom's Hardware: 1.7%Pessimism Bias0.0%This article: 17.0%Luke James: 5.8%Tom's Hardware: 6.3%Negativity Bias17.0%This article: 0.0%Luke James: 2.3%Tom's Hardware: 1.3%Self-Serving Bias0.0%This article: 6.1%Luke James: 0.6%Tom's Hardware: 0.6%Fundamental Attribution Error6.1%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.2%Actor-Observer Bias0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.5%In-Group Bias0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.2%Out-Group Homogeneity Bias0.0%This article: 0.0%Luke James: 1.7%Tom's Hardware: 2.9%Halo Effect0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%Horn Effect0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%Dunning-Kruger Effect0.0%This article: 1.3%Luke James: 2.5%Tom's Hardware: 1.9%Recency Bias1.3%This article: 0.0%Luke James: 0.8%Tom's Hardware: 0.6%Primacy Effect0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.1%Blind-Spot Bias0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.2%Ad Hominem0.0%This article: 0.0%Luke James: 0.1%Tom's Hardware: 0.1%Straw Man0.0%This article: 4.3%Luke James: 6.4%Tom's Hardware: 5.1%Appeal to Authority4.3%This article: 0.0%Luke James: 0.9%Tom's Hardware: 2.0%False Dilemma0.0%This article: 0.0%Luke James: 0.2%Tom's Hardware: 0.8%Slippery Slope0.0%This article: 0.0%Luke James: 0.1%Tom's Hardware: 0.1%Circular Reasoning0.0%This article: 6.1%Luke James: 4.0%Tom's Hardware: 5.7%Hasty Generalization6.1%This article: 0.0%Luke James: 0.4%Tom's Hardware: 0.2%Red Herring0.0%This article: 4.3%Luke James: 0.6%Tom's Hardware: 1.0%Bandwagon4.3%This article: 11.5%Luke James: 1.1%Tom's Hardware: 3.0%Appeal to Emotion11.5%This article: 5.7%Luke James: 0.8%Tom's Hardware: 0.8%Begging the Question5.7%This article: 4.3%Luke James: 2.7%Tom's Hardware: 3.4%Post Hoc (False Cause)4.3%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.1%Tu Quoque0.0%This article: 0.0%Luke James: 0.8%Tom's Hardware: 0.7%Burden of Proof0.0%This article: 0.0%Luke James: 0.1%Tom's Hardware: 0.3%Appeal to Nature0.0%This article: 0.0%Luke James: 0.6%Tom's Hardware: 0.5%Composition/Division0.0%This article: 7.4%Luke James: 1.0%Tom's Hardware: 1.7%Anecdotal7.4%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%No True Scotsman0.0%This article: 5.7%Luke James: 1.9%Tom's Hardware: 3.4%Ambiguity (Equivocation)5.7%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%Gambler’s Fallacy0.0%This article: 0.0%Luke James: 0.1%Tom's Hardware: 0.2%Middle Ground0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%Personal Incredulity0.0%This article: 5.7%Luke James: 0.1%Tom's Hardware: 0.2%Special Pleading5.7%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.1%Genetic Fallacy0.0%This article: 5.7%Luke James: 4.3%Tom's Hardware: 2.1%Unattributed Quote5.7%This article: 0.0%Luke James: 0.7%Tom's Hardware: 0.9%Quote-first Misdirection0.0%This article: 18.5%Luke James: 2.8%Tom's Hardware: 7.1%Biased Writer Voice18.5%This article: 0.0%Luke James: 0.3%Tom's Hardware: 1.3%Indoctrination0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.1%Politically Left Leaning Bias0.0%This article: 0.0%Luke James: 0.0%Tom's Hardware: 0.0%Politically Right Leaning Bias0.0%This article: 0.0%Luke James: 0.9%Tom's Hardware: 3.4%Attempt to Sell a Product or S…0.0%

460 words analyzed.

Speakers

1speaker5.7%attributed speech434writer words
Voice mapSelect a segment to jump to its words
Writer's voice • 27 words • 100.0% coverageWriter's voice • 58 words • 100.0% coverageWriter's voice • 16 words • 0.0% coverageWriter's voice • 41 words • 0.0% coverageWriter's voice • 33 words • 0.0% coverageWriter's voice • 36 words • 0.0% coverageWriter's voice • 39 words • 0.0% coverageWriter's voice • 26 words • 0.0% coverageWriter's voice • 11 words • 0.0% coverageWriter's voice • 6 words • 0.0% coverageWriter's voice • 34 words • 0.0% coverageWriter's voice • 26 words • 0.0% coverageWriter's voice • 20 words • 0.0% coverageWriter's voice • 28 words • 0.0% coverageWriter's voice • 13 words • 0.0% coverageTed Lieu • 26 words • 100.0% coverageWriter's voice • 20 words • 0.0% coverage
Selected voice

Ted Lieu

100%flagged-word coverage
26 attributed words100% of attributed speech83% writer coverage
0%50.0%100.0%Unattributed Quote+100.0 ptsWriter: 0.0%Ted Lieu: 100.0%100.0%Biased Writer Voice-19.6 ptsWriter: 19.6%Ted Lieu: 0.0%0.0%

Attribution is sentence-level. Pattern percentages are calculated only from words assigned to that voice.

Loading…
Loading…
Loading…

Analysis

Hover over highlighted words in the article to view the associated bias or fallacy analysis.