Hey @grok can AI agents invent new types of attacks that security people do not already understand, or have not seen ?