Anthropic

Not verified. Claim this page

Visit Website

Anthropic Sustainability Actions

Published responsible scaling policy

Anthropic published a Responsible Scaling Policy in September 2023 and later updated it in February 2026. The policy sets out how the company tries to reduce serious risks from more powerful AI models. It says stronger safeguards should be added as model capabilities rise. The company describes the policy as a guide for its own launches and training work, and as a model it hopes others in the AI sector will follow.

9 16

Built a frontier compliance framework

In December 2025, Anthropic published a Frontier Compliance Framework. This is a set of rules for checking and reducing risks from frontier models, meaning the most advanced AI systems. The framework covers cyber offense, chemical, biological, radiological, and nuclear threats, plus risks like sabotage, loss of control, and harmful manipulation. It also explains how the company protects model weights, which are the core settings that shape how a model works.

9 16

Earned ISO/IEC 42001

Anthropic says it achieved ISO/IEC 42001:2023, an international standard for managing AI systems. This standard is meant to help companies build and use AI in a more controlled and responsible way. The company presents this as outside validation of its AI management system and its risk framework. It is one of the few clearly named third-party standards mentioned in the material.

9 16

Uses red teaming and testing

Anthropic says it uses internal and external red teaming, which means structured attempts to find weaknesses before others do. It also runs regular evaluations, threat modeling, policy vulnerability testing, and pre-deployment checks before new models are released. The company says it works with outside groups such as the UK AI Security Institute, the US Center for AI Standards and Innovation, and METR to test models and study risks.

9 16

Set up safety reporting channels

Anthropic has created several ways for people to report safety problems. These include a responsible disclosure policy, bug bounty programs through HackerOne, and a safety email for users to report jailbreaks or other concerns. It also says employees can raise AI safety issues through internal channels, including an anonymous route. These systems are meant to catch problems after release and improve monitoring over time.

9 16

Started measuring emissions

A Fortune report in 2026 said Anthropic was working with Watershed, a carbon accounting platform, to measure its company-wide footprint and prepare for California reporting rules. The same report said the company had hired for sustainability and non-financial reporting roles. This suggests Anthropic has begun building the systems needed to track emissions, even though it has not publicly published a full sustainability report in the material provided.

13

Invested in energy research

In July 2025, Anthropic announced a $2 million contribution to Carnegie Mellon University. Half of the money supports energy research through the Scott Institute for Energy Innovation, and half supports a cybersecurity education program for middle and high school students. The company said the energy work could help improve grid management and support more efficient energy systems.

4 7 9 13

About Anthropic

  • Status
  • Unverified