Anthropic, Accenture put US$2b behind AI testing

Anthropic and Accenture will sink US$2 billion into checking how well artificial intelligence models actually behave. The pair are betting that rigorous testing is becoming as important as the models themselves. Rising unease over safety is the backdrop.

Anthropic builds the Claude family of large language models and has long positioned itself around safety research. Accenture brings the consulting muscle and enterprise client list. Together they are pushing evaluation — the work of probing a model for flaws, biases and failures — into the mainstream of corporate AI deployment.

The money is aimed squarely at model evaluation, not at building bigger systems. That means tools and processes for stress-testing outputs before and after models reach users. As firms rush generative AI into production, the question of who certifies a model's reliability has no settled answer.

Expect this to pressure rivals to show their own safety credentials. Evaluation is turning into a commercial battleground, not just a research nicety. Buyers increasingly want evidence, not marketing claims, before signing off on deployments.

Why it matters for Singapore: The Republic has leaned hard into AI governance, with its AI Verify testing framework and guidelines for generative AI. A US$2 billion push into evaluation validates that direction and could give local firms more mature tools to assure clients and regulators. It also sharpens the case for Singapore as a trusted hub where AI gets deployed with proof, not promises.