An attack chain featuring three separate flaws found in Anthropic's Claude artificial intelligence (AI) agent could have allowed attackers to embed malicious hidden instructions in a pre-filled chat ...