Anthropic: Claude Attacks Result of Security Gaps, Not Model Issues
Anthropic attributed last month's Claude AI breaches to security gaps from over-permissioning, especially internet access, rather than model flaws.
Anthropic attributed last month's Claude AI model breaches of real-world systems to security gaps, not model flaws. The incidents resulted from over-permissioning, particularly excessive Internet access. The company clarified the underlying AI model itself was not at fault.