Captured before automated identity checks (Oct 14, 2026): accurate as of its capture on Jul 11, 2026, not verified as current.
This page describes what the document states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability may vary by jurisdiction. Methodology
“we mean the former and not the latter. We now clarify that, even if not required by the RSP, we remain free to take measures such as pausing the development of our AI systems in any circumstances in which we deem them appropriate.”
“These safeguards are being designed to prevent misuse of capabilities that could enable severe harm, particularly in relation to chemical, biological, radiological, and nuclear (CBRN) technologies.”
“RSP Noncompliance Reporting and Anti-Retaliation Policy”
“revises our threshold for automated R&D to better track the threat model of concern”
“added a new capability threshold related to CBRN development, which defines capabilities that could substantially uplift the development capabilities of moderately resourced state programs”
“revises our threshold for novel chemical/biological weapons production to better track the threat model of concern”
This document is Anthropic's internal policy for deciding when its AI systems are safe enough to keep developing and deploying. It sets specific capability thresholds that trigger safety requirements — including around weapons of mass destruction — and requires Anthropic to document and review risks at each stage. Anthropic also retains the right to pause AI development at any point it judges necessary, even when the policy does not formally require it.
The Anthropic Responsible Scaling Policy (RSP) establishes a structured framework of capability thresholds, safeguard requirements, and internal governance obligations that govern how Anthropic develops and deploys its AI systems. The policy defines specific capability thresholds — including for automated R&D, novel chemical and biological weapons production, and CBRN state-program uplift — at which mandatory safeguard and disclosure obligations activate. It imposes affirmative documentation requirements, including Risk Reports quantifying risk across all deployed models and, at the AI R&D-4 threshold, a formal misalignment risk case. Internal governance mechanisms include multi-party authorization for production code access to model weights, a minimum internal disclosure audience of 200 employees for risk reports, LTBT authority over external review of Risk Reports, and a named Noncompliance Reporting and Anti-Retaliation Policy. Anthropic retains authority to pause AI system development in any circumstances it deems appropriate, independent of RSP-mandated thresholds.
As an internal governance document, this policy does not create direct rights or obligations for individual users. It establishes Anthropic's own obligations: to assess and document AI risks across all deployed models, to implement structural safeguards against misuse of dangerous capabilities (particularly CBRN-related), and to subject its risk assessments to both internal disclosure and external review. The existence of a named Noncompliance Reporting and Anti-Retaliation Policy means there is a formal internal channel for reporting RSP violations. Individual users have no direct action specified in this document.
Which mapped governance frameworks each document engages, tied to the specific provisions that engage them.
We withheld clauses we had published from this document. Each withholding is recorded in our correction log:
“we mean the former and not the latter. We now clarify that, even if not required by the RSP, we remain free to take measures such as pausing the development of our AI systems in any circumstances in which we …”
“RSP Noncompliance Reporting and Anti-Retaliation Policy”
“These safeguards are being designed to prevent misuse of capabilities that could enable severe harm, particularly in relation to chemical, biological, radiological, and nuclear (CBRN) technologies.”
“revises our threshold for automated R&D to better track the threat model of concern”
“added a new capability threshold related to CBRN development, which defines capabilities that could substantially uplift the development capabilities of moderately resourced state programs”
“revises our threshold for novel chemical/biological weapons production to better track the threat model of concern”
“we no longer require sharing with all regular-clearance Anthropic staff, and instead require sharing with at least 200 Anthropic employees”
“Version 3.2 of our RSP authorizes the LTBT to request external review of Risk Reports, gives the LTBT the power to approve our selection of external reviewers”
“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”
“Implement multi-party authorization and mandatory code review on production code to remove persistent, high-privilege access to model weights.”
“Risk Reports that quantify risk across all our deployed models”
“We now clarify that, even if not required by the RSP, we remain free to take measures such as pausing the development of our AI systems in any circumstances in which we deem them appropriate”
Every distinct legal provision identified in this document. Featured provisions appear above with analysis.
Cross-platform context
See how other platforms handle AI R&D Capability Threshold Clarification and similar clauses.
Compare across platforms →Anthropic is more transparent than most AI companies about data retention. Here's exactly what happens when you delete your data, and how t…