Anthropic starts embedded evaluations with Accenture-backed Faculty getting near-employee access

Anthropic starts embedded evaluations with Accenture-backed Faculty getting near-employee access

N
News Editor
2026-09-19 02:28:32
Anthropic has begun rolling out what it calls "embedded evaluations," a model assessment setup that gives outside evaluators access much closer to that of internal staff than a standard third-party review. The first participant is Faculty, an AI company under Accenture, whose personnel will work inside Anthropic to evaluate models, run red-team tests, conduct alignment reviews, and examine safety measures. According to BlockBeats, evaluators in this arrangement can view parts of the model training process, learn how development and deployment decisions are made, and communicate directly with Anthropic employees. Anthropic and Accenture plan to invest at least $1 billion each over the next five years to build out this evaluation capability. Anthropic also said the industry still lacks a common rulebook for this kind of work. Key questions remain unsettled, including what evaluators should be allowed to see, how findings should be disclosed, and who should pay for the process. For now, Anthropic is directly paying for Accenture's evaluation work. That has drawn attention because Accenture is already a long-term strategic partner of Anthropic and has created an Anthropic Business Group that has trained about 30,000 people to use Claude. On the same day, the AI Evaluator Forum said one minimum condition for credible evaluation is that evaluators should not have major commercial ties with the company being assessed.

Anthropic has formally started rolling out what it calls "embedded evaluations," with Faculty, an AI company under Accenture, becoming the first outside group to enter the company under the new setup, according to BlockBeats.

Faculty staff will work inside Anthropic to evaluate models, run red-team tests, conduct alignment reviews, and check safety measures. The arrangement goes well beyond a standard external review.

Evaluators will receive access close to that of employees

Under the model, evaluators will be given permissions close to employee level. That includes the ability to view the model training process, understand how development and deployment decisions are made, and communicate directly with internal staff.

That makes the program materially different from a typical outside assessment, where reviewers usually test systems from the perimeter rather than from inside the organization.

Anthropic and Accenture each plan to commit at least $1 billion

The two sides plan to invest at least $1 billion each over the next five years to build this evaluation capability.

Anthropic said the sector still does not have a unified set of rules for how such evaluations should work. Questions including what evaluators can see, how results should be made public, and who should cover the cost are still being worked out.

Commercial ties raise independence questions

For now, Anthropic is directly paying for Accenture's evaluation work. At the same time, Accenture is already a long-term strategic partner of Anthropic. It has also set up an Anthropic Business Group that has trained about 30,000 people to use Claude.

That point stood out because, on the same day, the AI Evaluator Forum said one of its minimum conditions is that an evaluation organization should not have major commercial dealings with the company it is assessing.

More evaluators may be added later

Anthropic plans to bring in other evaluation organizations after Faculty. It is also discussing pilot programs with nonprofit groups including METR, which BlockBeats described as another longtime Anthropic partner.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
1100

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.