Anthropic updates its Responsible Scaling Policy (v2.0): Capability Thresholds and Required Safeguards
In one sentence On October 15, 2024 Anthropic publishes version 2.0 of its Responsible Scaling Policy: capability thresholds for CBRN weapons and autonomous AI research, tied to ASL safety standards, new capability and safeguard assessments, and Jared Kaplan as Responsible Scaling Officer.
Anthropic, the company that builds Claude, has a public policy setting out when a model becomes capable enough to need stronger protections. It is called the Responsible Scaling Policy, and an updated version comes out on October 15, 2024.
The core commitment stays the same: do not train or deploy models without safety measures that keep risks at an acceptable level. What is new is more precise capability thresholds: if a model can meaningfully help someone with a basic technical background build chemical, biological, radiological or nuclear weapons, or if it can carry out complex AI research on its own, stricter protections apply. At the time of the update all Anthropic models operate under the ASL-2 standard.
Companies
Anthropic
Tools
—
Tags
Sources