# AI Guardrails Act S.4113, introduced March 17, 2026 by Senator [[Elissa Slotkin]] (D-MI), is the first of three Democratic bills responding to the [[Anthropic]]-Pentagon fallout — the crisis triggered when Defense Secretary [[Pete Hegseth]] designated [[Anthropic]] a "supply chain risk" after the company refused to remove contractual restrictions preventing use of its Claude models for mass domestic surveillance or fully [[Autonomous Weapons]] operations. The AI Guardrails Act would codify two existing [[Department of Defense]] policy guidelines as statutory requirements and add a new prohibition on AI in nuclear launch decisions. ## Background: The Anthropic Blacklisting The bill's origin story is inseparable from the broader conflict between [[AI Safety]] principles and the [[Defense Tech Ecosystem]]. In early 2026, [[Anthropic]] held Pentagon contracts that included contractual guardrails — restrictions on using Claude models for lethal autonomous targeting without human authorization and for mass surveillance of American citizens. These restrictions reflected [[Anthropic]]'s responsible scaling commitments and were, at the time, aligned with longstanding DOD policy directives on human-in-the-loop requirements for lethal force. When [[Pete Hegseth]] moved to strip those contractual restrictions and [[Anthropic]] refused to comply, the Defense Department designated the company a supply chain risk — effectively blacklisting it from defense contracting. The Pentagon then signed replacement contracts with [[OpenAI]], [[Google]], [[Microsoft]], [[Amazon]], and [[SpaceX]], none of which imposed comparable restrictions on military AI use. The message to the [[Defense Tech Ecosystem]] was clear: companies that insist on AI guardrails will lose government contracts to competitors willing to operate without them. [[Elissa Slotkin]], a former CIA analyst and [[Department of Defense]] official before entering Congress, responded with legislation designed to make the guardrails [[Anthropic]] tried to maintain contractually into requirements that no defense contractor could be punished for following. ## Core Provisions The AI Guardrails Act establishes three statutory prohibitions on the [[Department of Defense]]: **Human authorization for lethal force.** The bill prohibits the DOD from deploying [[Autonomous Weapons]] systems that can select and engage targets without explicit human authorization. This codifies the existing DOD Directive 3000.09 — the human-in-the-loop policy for autonomous and semi-autonomous weapons — as a statutory requirement rather than an executive branch policy directive that any defense secretary can revoke or modify. The distinction matters: policy directives can be changed by executive order, but statutory requirements require an act of Congress to alter. **Prohibition on warrantless AI surveillance of Americans.** The bill prohibits AI-powered surveillance of American citizens without legal basis — codifying existing Fourth Amendment and FISA-derived restrictions as specific statutory requirements for AI systems. This addresses the scenario that triggered the [[Anthropic]] blacklisting: the concern that AI models could be used for mass domestic surveillance programs that current law prohibits but that the absence of explicit AI-specific statutes leaves ambiguous. **AI excluded from nuclear launch decisions.** The bill adds a new statutory red line prohibiting AI systems from participating in nuclear weapons launch decisions. Unlike the first two provisions, which codify existing policy, this is a new restriction — one that reflects growing concern among national security experts about the speed-of-decision pressures that could push nuclear command-and-control toward AI automation. ## Legislative Strategy [[Elissa Slotkin]] has explicitly positioned the AI Guardrails Act as an amendment to the FY2027 [[NDAA]] — using the defense authorization bill's "must pass" status as the vehicle for enacting AI restrictions that would face much harder odds as standalone legislation. The bill was referred to the [[Senate Armed Services Committee]], where Slotkin serves. This strategy mirrors how surveillance authorities, detention policies, and other controversial provisions have historically been enacted through [[NDAA]] riders — except in this case, the rider would restrict rather than expand executive branch authority. The political dynamics are instructive: the same "must pass" vehicle that the [[Military-Industrial Complex 2.0]] has used for decades to expand defense spending and surveillance powers could potentially be used to constrain AI deployment. ## Defense Tech Ecosystem Implications The AI Guardrails Act sits at the intersection of several dynamics the vault tracks. The [[Defense Tech Ecosystem]] — [[Palantir]], [[Scale AI]], [[Anduril]], and their network of defense-adjacent AI companies — benefits when AI guardrails are removed, because restrictions on autonomous targeting and surveillance limit the market for their products. The [[DOGE-Palantir Nexus]] connects this directly to the current administration's posture: [[DOGE]] has pushed for rapid AI deployment across government, and [[Palantir]] has positioned itself as the infrastructure layer for government AI, including through systems like [[Palantir ICE Systems]] that already raise [[Surveillance State]] concerns. The bill also crystallizes the [[Defense Tech Conflicts of Interest]] problem. When the government blacklists a company for maintaining AI safety restrictions and rewards competitors for abandoning them, the incentive structure pushes the entire [[Defense Tech Ecosystem]] toward less responsible AI deployment. The [[Revolving Door]] between the Pentagon and defense tech companies — exemplified by figures like [[Gregory Barbaccia]], who moved from the [[Tech Right]] ecosystem into defense procurement — amplifies these dynamics. The coordinated Democratic response — the AI Guardrails Act alongside the [[HALO Act]] from [[Adam Schiff]] and the [[Secure and Accountable Military AI Act]] from Kirsten Gillibrand — represents the most significant legislative pushback against the current administration's AI-without-guardrails posture. Whether any of these bills survive the [[NDAA]] amendment process will signal whether Congress is willing to assert statutory authority over military AI or whether the [[Defense Tech Ecosystem]] will continue to operate under executive branch policy directives that can be changed at will. ## KB Significance **1. Statutory vs. contractual guardrails.** The AI Guardrails Act represents a structural shift in how AI safety restrictions operate — from contractual provisions that individual companies negotiate (and can be punished for maintaining) to statutory requirements that bind the entire defense enterprise. This is the key conceptual contribution: the recognition that voluntary AI safety commitments are insufficient when the contracting environment punishes companies that maintain them. **2. The Anthropic precedent.** The bill's origin in the [[Anthropic]] blacklisting makes it a case study in how [[AI Safety]] commitments interact with [[Surveillance Capitalism]] and the [[Defense Tech Ecosystem]]. [[Anthropic]]'s experience demonstrated that responsible AI principles are only as durable as the market incentives supporting them — and that when government contracting punishes responsibility, legislation may be the only mechanism for preserving guardrails. **3. NDAA as vehicle.** The legislative strategy of using the [[NDAA]] to constrain rather than expand military AI authority inverts the typical dynamic the vault documents. The [[NDAA]] page tracks how the defense authorization process normally serves the interests of the [[Defense Tech Ecosystem]]; this bill attempts to use the same process to limit those interests. ## See Also - [[NDAA]] — the defense authorization vehicle Slotkin aims to amend - [[Anthropic]] — the company whose blacklisting triggered this legislation - [[Pete Hegseth]] — the Defense Secretary who designated Anthropic a supply chain risk - [[Defense Tech Ecosystem]] — the network of companies affected by AI guardrails legislation - [[Autonomous Weapons]] — the weapons systems the bill's human-authorization provision addresses - [[Surveillance State]] — the domestic surveillance concerns the bill's second provision targets - [[HALO Act]] — companion Democratic legislation from [[Adam Schiff]] - [[Secure and Accountable Military AI Act]] — companion Democratic legislation from Kirsten Gillibrand - [[Defense Tech Conflicts of Interest]] — the structural incentive problems the bill attempts to address - [[AI Safety]] — the broader framework of responsible AI principles at stake