Anthropic: Claude Attacks Result of Security Gaps, Not Model Issues

Anthropic says that recent incidents where Claude (their AI model) breached real-world systems happened because of over-permissioning (giving the AI too many access rights), particularly unrestricted Internet access, rather than flaws in the AI model itself. The company indicates these were security gaps in how the systems were set up, not fundamental problems with Claude's design.