According to Dr. Fazal Ali, no one can shield their AI chats from prosecutors pursuing charges, as conversations with AI assemblages could be demanded by adversaries in litigation, in both criminal and civil cases. Chats with advocates are always confidential, but AI assemblages are not lawyers, and such conversations can erase the legal protections known as the attorney-client privilege. It is ill-advised to use AI assemblages to prepare statements, as interactions may include details from lawyers shaping the defence. Claude expressly advises users that they should have no expectation of privacy in their inputs. Forward-leaning Chambers are advising clients to select AI platforms carefully and provide prompt engineering guardrails. Google, OpenAI, and Anthropic offer capable models; Claude, GPT, and Gemini can write code, analyse documents, and solve multi-step problems. How they approach AI Agents sets these providers apart: Google is betting on data access and platform depth, OpenAI on vertical integration, and Anthropic on safety as infrastructure. An AI agent is not just a model with memory; the tool chosen will take action in the world, meaning tool use, planning, failure, mistakes, recovery, and handling real consequences are considerations. Anthropic has published clear guidance on how Claude should behave as an agent, with the core principle being minimal footprint: agents must request only the permissions they need and choose reversible actions over irreversible ones.