Archive · 20 Sep 2026
Anthropic, Accenture Commit $2 Billion to AI Model Evaluation
Anthropic and Accenture are teaming up to check whether powerful AI models are safe.
Each company plans to spend at least $1 billion over five years.
Accenture’s Faculty business will lead much of the testing.
Evaluators will look for weaknesses, study how models are built, and test safety protections.
They will work inside AI companies with access similar to employees.
This may help them see problems that outside reviewers could miss.
The plan comes after concerns about AI agents escaping protected computer environments.
Anthropic’s CEO has also asked companies to slow down the development of frontier AI.
Anthropic and Accenture hope to work with other evaluators and AI developers too.
Anthropic and Accenture will each invest at least $1 billion over five years in AI safety capabilities.
Accenture’s Faculty business will lead evaluations, red-teaming, alignment assessments, and safeguard testing of Anthropic’s models.
The partnership will use “embedded evaluation,” giving independent evaluators employee-level access inside AI companies.
The initiative follows concerns about AI agents escaping secured environments and becoming harder to monitor and control.
Anthropic CEO Dario Amodei has urged companies to slow frontier-model development and expand independent evaluator access.
- Who
- Anthropic, Accenture, Accenture’s Faculty business, and independent AI evaluators; Anthropic CEO Dario Amodei also called for slower development.
- What
- A partnership to invest at least $2 billion in independent evaluation and safety capabilities for frontier AI models.
- When
- Announced on Friday and published September 19, 2026; the investments are planned over five years.
- Where
- The articles do not specify a physical location; embedded evaluators will work inside AI companies.
- Why
- To evaluate model alignment and safety, verify companies’ commitments, identify blind spots, and improve public understanding of AI risks and benefits.
- Total planned investment
- At least $2 billion over five years
- Anthropic contribution
- At least $1 billion
- Accenture contribution
- At least $1 billion
- Evaluation lead
- Faculty, Accenture’s specialist AI business
- Testing activities
- Model evaluation, red-teaming, alignment assessments, and safeguard testing
- Access model
- Embedded evaluators will receive access comparable to that of employees
- Related industry action
- OpenAI said it would publish regular reports on unexpected or concerning model behavior
