Print-friendly view · use your browser's Save as PDF option (Cmd/Ctrl-P) to attach this article to a brief.

AI Safety Level 3 (ASL-3)

asl-3 · safety · concept

Source: https://policywindow.org/wiki/asl-3

Generated 2026-05-30T22:10:25 UTC

Summary

A capability-based risk tier in Anthropic's Responsible Scaling Policy denoting models with the potential to substantially uplift CBRN attack capabilities or autonomous AI replication.

At a glance

Used by: 3 instrument(s)
Related concepts: frontier-tier, systemic-risk, compute-threshold
Primary source: Anthropic Responsible Scaling Policy v1.x
Source URL: https://www.anthropic.com/news/anthropics-responsible-scaling-policy

Details

ASL-3 was introduced in Anthropic's Responsible Scaling Policy (RSP) framework. Triggering ASL-3 capability requires the model to demonstrate substantial uplift in chemical, biological, radiological, or nuclear (CBRN) weapons design beyond baseline internet resources, OR show signs of autonomous self-replication. ASL-3 status mandates specific deployment safeguards including red-team evaluations, restricted API access, and incident-response protocols. Comparable tiers exist in OpenAI's Preparedness Framework (high) and DeepMind's Frontier Safety Framework (Critical Capability Levels).

How to cite this article

APA

Policy Window. (n.d.). AI Safety Level 3 (ASL-3) [Wiki article — Concept]. https://policywindow.org/wiki/asl-3

Chicago

Policy Window. n.d.. "AI Safety Level 3 (ASL-3)." Wiki article (Concept). https://policywindow.org/wiki/asl-3.

Harvard

Policy Window (n.d.) 'AI Safety Level 3 (ASL-3)', Wiki article — Concept, available at: https://policywindow.org/wiki/asl-3.

OSCOLA

Policy Window, 'AI Safety Level 3 (ASL-3)' (Wiki article — Concept, n.d.) <https://policywindow.org/wiki/asl-3> accessed [date].

BibTeX

@misc{policywindow-asl-3,
  title  = {AI Safety Level 3 (ASL-3)},
  author = {Policy Window},
  year   = {n.d.},
  howpublished = {asl-3 — safety},
  url    = {https://policywindow.org/wiki/asl-3},
  note   = {Primary source: https://www.anthropic.com/news/anthropics-responsible-scaling-policy}
}