Anthropic has chosen Accenture to assess the safety of its models

Anthropic has entered into an agreement with consulting firm Accenture, which will become the first external partner to evaluate the safety and reliability of the models being developed. As part of the collaboration, the company's specialists will conduct an independent audit of systems before their public release.

The program involves embedding experts in testing processes, where they will check models for resilience to cyber threats, bias, and compliance with ethical standards. This is the first case where a developer like Anthropic delegates the functions of an "embedded evaluator" to a third-party organization working with corporate clients.

During the checks, closed datasets and specific attack scenarios developed to identify vulnerabilities in the LLM architecture will be used. The audit results will be taken into account when making decisions about releasing updates to end-users and the business segment.

The collaboration is aimed at reducing the risks associated with deploying AI solutions in critical industries such as finance and healthcare. Previously, such checks were mainly carried out by internal security teams, but the growing complexity of models required the involvement of external experts for independent verification.

This step reflects the general trend in the industry, where companies are striving to increase the transparency of their developments through external partnerships. The use of third-party auditors allows Anthropic to confirm the reliability of its systems for large corporate customers, who require strict security guarantees before implementing technologies into their workflows.