Microsoft’s multi-agent AI system tops Anthropic’s Mythos on cybersecurity benchmark - GeekWire — AI News