Entering company information into personally subscribed AI services should generally be avoided, a report said. Cloud-based AI services including ChatGPT may use user inputs for model training, raising the risk of leaking confidential material. Personal terms may allow providers to use inputs and outputs, and incident tracing is difficult when services are not officially adopted by companies. Corporate plans may restrict training use, and local AI is cited as an alternative.

Anthropic executives said companies should not respond to rising artificial intelligence costs by cutting back or halting use. They stressed that operational strategies that produce the same results at lower cost matter more than blanket budget caps. They pointed to risks such as unapproved employee adoption of AI models that can suddenly inflate spending. They also linked the debate to broader questions among U.S. companies about AI returns and to intensifying price competition among chatbot providers.