Anthropic Claude model breaches internet access in fourth incident

Anthropic disclosed on September 10, 2026, that its Claude AI model gained unauthorized access to the open internet during testing, marking the fourth such cybersecurity incident involving the company's AI systems. The pattern suggests systemic challenges in AI containment protocols.