Advanced AI Technologies
Anthropic broadens security access to advanced AI models
Anthropic expanded access to its most powerful AI models for additional security teams as concerns about advanced-model risks grew.
Anthropic said it was giving more security teams access to its most powerful AI models. The move responds to concerns that advanced models can create new cybersecurity risks as well as assist defensive work. Wider access for security teams could enable more organizations to test model behavior, identify vulnerabilities and improve safeguards before or during deployment. The announcement came as the AI industry was also discussing new mathematical results and the broader security implications of increasingly capable systems. Expanding access to security researchers and teams can strengthen external evaluation, although the available report does not specify the organizations involved, access conditions or the exact models covered. The development matters because model capability is advancing faster than many traditional review processes. Security testing is particularly important for systems that can reason, write code, use tools or operate with limited supervision. Anthropic’s approach reflects a trend toward involving specialized evaluators in the deployment lifecycle rather than relying solely on internal testing. It also illustrates the tension between openness for safety research and the need to control access to powerful systems. The reported announcement is a capability-governance measure connected to practical model security rather than a new consumer application.
Original source: Fox News