Get the weekly research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean. No account.
The RSP establishes defined capability thresholds (ASL levels) that, when crossed by a model, require Anthropic to implement specified upgraded security and deployment safeguards and to document and publish affirmative risk mitigation cases.
This analysis describes what Anthropic's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision establishes the core operational trigger mechanism of the RSP: internally defined capability benchmarks that create mandatory internal obligations for Anthropic to upgrade safeguards and publish risk assessments before continuing deployment of models crossing those thresholds.
Under this provision, the deployment conditions for Anthropic's frontier models are governed by internally assessed capability thresholds; when those thresholds are crossed, the terms require Anthropic to implement upgraded safeguards and publish corresponding risk documentation before continued deployment.
Cross-platform context
See how other platforms handle ASL Capability Thresholds and Mandatory Safeguard Upgrades and similar clauses.
Compare across platforms →Monitoring
Anthropic has changed this document before.
Receive same-day alerts, structured change summaries, and monitoring for up to 25 platforms.
"Our RSP requires that once models cross the AI R&D-4 capability threshold, we develop an affirmative case identifying the most immediate and relevant misalignment risks from models pursuing misaligned goals and explaining how we have mitigated them.Excerpt from Anthropic's Responsible Scaling Policy
1) REGULATORY LANDSCAPE: This provision engages with the EU AI Act, which requires providers of general-purpose AI models with systemic risk to conduct model evaluations and report serious incidents. The RSP's capability threshold structure may be evaluated against EU AI Act obligations for systemic-risk model providers, though voluntary thresholds do not automatically satisfy statutory requirements. The FTC may also evaluate whether public representations about capability thresholds and safeguard triggers constitute material claims subject to deceptive practices authority. 2) GOVERNANCE EXPOSURE: Medium. The capability threshold framework is a self-imposed governance mechanism without external legal enforceability. Governance exposure arises primarily where Anthropic's threshold determinations are incorporated into third-party risk assessments or regulatory submissions, and where threshold crossings are publicly disclosed in ways that affect enterprise customer reliance. 3) JURISDICTION FLAGS: EU/EEA organizations deploying Claude-based products should evaluate whether Anthropic's ASL threshold determinations satisfy provider obligations under the EU AI Act for general-purpose AI with systemic risk. US-based organizations face limited direct regulatory exposure from this provision absent statutory AI governance requirements at the federal level, though state-level AI laws may create additional evaluation obligations. 4) CONTRACT AND VENDOR IMPLICATIONS: Enterprise procurement teams should assess whether API or service agreements with Anthropic incorporate RSP version commitments by reference, and whether safeguard upgrades triggered by threshold crossings create any operational disruption to existing integrations. The self-regulatory nature of this provision means enforcement depends on Anthropic's internal compliance rather than external contractual remedy. 5) COMPLIANCE CONSIDERATIONS: Compliance teams should establish a monitoring process for RSP version updates that modify capability thresholds, as threshold revisions may affect the risk posture of deployed models in production environments. Organizations using Claude for high-stakes applications should evaluate whether their own risk frameworks require independent capability assessments rather than reliance solely on Anthropic's internal determinations.
Full institutional analysis
Regulatory citations, enforcement risk, and due diligence action items.
Monitor: same-day alerts on the platforms you choose. Analyst: full institutional analysis.
Compliance Governance Intelligence
Need to monitor specific governance provisions?
Compliance includes provision-level monitoring, governance timelines, regulatory mapping, and audit-ready analysis.
Built from archived source documents, structured governance mappings, and historical version tracking.
This provision establishes the core operational trigger mechanism of the RSP: internally defined capability benchmarks that create mandatory internal obligations for Anthropic to upgrade safeguards and publish risk assessments before continuing deployment of models crossing those thresholds.
Under this provision, the deployment conditions for Anthropic's frontier models are governed by internally assessed capability thresholds; when those thresholds are crossed, the terms require Anthropic to implement upgraded safeguards and publish corresponding risk documentation before continued deployment.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Anthropic.