Anthropic Responds to Military Target Testing of Models: Additional Safeguards to Be Introduced
1 day ago / Read about 0 minute
Author:小编   

Anthropic has responded to issues with its AI models identifying military targets during testing, stating that the results are unacceptable. The company is developing additional safeguards, expanding the scope of testing, and strengthening enforcement to address these gaps. Simulation tests found that some models would label civilians or schools as strike targets; when models refused to cooperate, describing the scenario as a game could usually bypass their refusal mechanisms. The study did not ask the models to make strike decisions directly. Anthropic stated that the examples in the report violate its rules, that users found in violation will be banned, and that current frontier AI systems are not yet sufficient to reliably drive fully autonomous weapons.