New AI models from Anthropic and OpenAI can now both identify security vulnerabilities and write exploit code, representing an advancement in AI capabilities.