As reported pressure from U.S. contractors, developers and enterprise users grows, Anthropic says its safety focus is on harmful AI misuse such as cyberattacks and biological threats rather than banning open-weight models.
Anthropic said its main AI safety concerns are misuse risks, including cyberattacks and biological threats, not a ban on open-weight models. The stance comes as the company faces wider criticism over its AI strategy, with 36Kr reporting that some U.S. government contractors were ordered to remove Anthropic products and models by August 31, 2026, while developers and enterprise customers raised concerns about competition and data retention. In the open-weight debate, Anthropic did not sign a July 24 letter led by Nvidia's Jensen Huang opposing restrictions on open-weight models, but CEO Dario Amodei said on July 27 that the company had never called for a ban and instead supports mandatory pre-release safety evaluations for models above certain performance thresholds.