9.5 Professional Ethics Considerations and Practice Guidelines in AI Applications

The Ethical Compass in the Intelligent Era: Practical Guidelines for Legal Professional Ethics in AI Applications

In the legal profession, where trust, justice, confidentiality, and professional judgment are lifelines, introducing any new technology that could alter its functioning must undergo extremely rigorous and prudent ethical scrutiny. Artificial intelligence (AI), with its unprecedented information processing power, potential for (even limited) autonomous decision-making tendencies, and unavoidable “black box” characteristics and risk vulnerabilities, places a series of classic legal professional ethics issues into new, more complex, and sometimes more ambiguous contexts.

Merely understanding general AI ethics principles (see Section 6.3) or being aware of potential risks in AI applications (see Sections 6.1, 6.6, etc.) is far from sufficient. The real challenge lies in how legal professionals can internalize these ethical considerations and externalize them into actions in every specific aspect of their daily work—from client intake, research, document drafting, to client communication, colleague collaboration, and even court appearances—maintaining constant vigilance and making responsible choices that meet the highest professional standards.

This section aims to provide more operational, practice-oriented guidance and a series of self-check checklists, acting as an “ethical compass” to help lawyers, in-house counsel, and other legal practitioners navigate the ethical fog, identify potential pitfalls, and firmly adhere to professional bottom lines when applying AI technologies in specific scenarios.

Core Guiding Principle: Technology always serves ethics; efficiency always yields to responsibility.

Checklist 1: Safeguarding the Duty of Confidentiality - The Cardinal Duty of Legal Professionals

Applicable Scenarios: Any use of AI tools to process tasks that might involve (however minimally or indirectly) client information. This includes, but is not limited to: using LLMs for legal research (when inputting case-related queries or background), using AI contract review tools (uploading contract drafts), using speech-to-text services (uploading recordings containing client conversations), using AI to assist drafting legal documents (when content is based on specific client information), interacting with potential clients via AI chatbots, etc.

Consideration Point	Practical Guidance & Self-Check	√/X
Tool Selection & Security Review	□ Do I only use AI tools or platforms that have been formally approved by my organization (firm/legal dept.) and have passed rigorous security and compliance reviews for handling any client-related information? □ Have I personally read carefully and fully understood the Privacy Policy, Data Processing Addendum (DPA), and Terms of Service (ToS) of the AI tool I am using? □ Does the tool provider clearly and unambiguously commit in writing that they will never use my input data (prompts, uploads) to train or improve their general AI models? Is there a clear, easy-to-use, default-on “Opt-out” mechanism regarding data use for training? □ Does the tool’s data storage location (involving cross-border transfer?), encryption measures (in transit and at rest), and access control mechanisms comply with the data protection regulations in my jurisdiction (e.g., PIPL, GDPR) and my organization’s internal security standards? □ For handling top-level sensitive or privileged information, have I prioritized solutions offering complete local deployment (e.g., running open-source models on internal servers) or enterprise-grade dedicated solutions with the highest level of data isolation and security guarantees?
Prudent Handling Before Data Input	□ Before inputting any content containing client information into an AI tool, have I prudently assessed: Is it necessary to input this information? Can the information be maximally anonymized or masked without significantly impacting the task outcome? (e.g., replacing client names with codes, redacting specific amounts/dates). □ If inputting original text with sensitive info is necessary (e.g., reviewing contract clauses), have I thoroughly removed or replaced all identifiable client information (contact details, addresses) and other sensitive details irrelevant to the current task? □ Am I adhering to the data minimization principle, inputting only the minimum amount of information necessary to complete the specific task at hand?
Transparency & Consent with Clients	□ Before planning to substantially use AI tools to process a client’s significant information or case materials (especially non-public info), have I proactively and clearly explained to the client that we may use AI assistance, outlining the main tools involved, general purpose, potential risks (even if minor), and the core measures we take to protect their information security and confidentiality? □ Depending on the circumstances (e.g., highly sensitive data, potential privilege waiver risks, explicit client request), have I obtained the client’s explicit, written informed consent?
Strict Internal Management & Access Control	□ Have I ensured that within my organization, access permissions to AI tools and client-related data processed by AI are strictly limited to the minimum scope of personnel with a genuine “Need-to-Know”? □ Are there complete, immutable operational audit logs recording the usage of AI tools and access to related data, facilitating security audits and accountability tracing?
Utmost Vigilance with Privileged Information!	□ Do I deeply understand the importance of attorney-client privilege and work product protection, and do I in principle, strongly avoid inputting any core communications or mental impression materials clearly falling, or highly likely to fall, under privilege protection into any third-party AI system (i.e., systems not fully controlled by the organization)? □ If processing is deemed absolutely necessary in extremely exceptional, fully justified, and approved circumstances, have I ensured the highest level of security isolation measures are in place, and have the potential privilege waiver risks been rigorously assessed and documented?

Checklist 2: Ensuring Professional Competence - New Requirements in the AI Era

Applicable Scenarios: When deciding whether and how to use AI tools to assist with various legal tasks (e.g., legal research, case analysis, document drafting or review).

Consideration Point	Practical Guidance & Self-Check	√/X
Understanding of AI Tool Used	□ Do I have at least a basic understanding of how the AI tool I am using or plan to use works? (e.g., What type of technology is it based on? Rule engine, statistical model, deep learning LLM? Is it a general foundation model or one optimized for a specific legal domain?) □ Do I clearly recognize the capability boundaries of this tool? What tasks is it good at? Where does it have known limitations or weaknesses? (e.g., What is its knowledge cutoff date? On which types of legal questions is it prone to “hallucinations” or errors? How accurate is it with long texts in my language or specific accents?) □ Have I spent time reading the tool’s official documentation, user manuals, or reliable third-party reviews and best practice guides?
Mastery of Necessary Skills	□ Do I possess the essential skills needed for effective and efficient interaction with this AI tool (especially if it’s an LLM)? For instance, do I understand the basic principles and common techniques of prompt engineering? (Ref Part 4) □ Do I know how to critically interpret and evaluate the quality, reliability, and potential risks of the outputs generated by this tool? (Ref Section 6.4)
Sufficient Awareness of Risks	□ Am I fully aware of the main potential data security risks, client privacy breach risks, algorithmic bias risks, information inaccuracy (hallucination) risks, intellectual property risks, and related legal compliance risks associated with using this AI tool? □ Am I clear on the initial response measures to take and the internal reporting process (which department/person) if I encounter problems or identify risks?
Wise Judgment on When/Not to Use	□ Before deciding to use AI for a specific task, have I rationally assessed: Is this task truly suitable for AI assistance? Does using AI genuinely offer more efficiency, better results, or unique value compared to traditional manual methods? (Avoid using AI just for the sake of using AI). □ For core legal tasks that are extremely complex, require deep creative thinking, involve significant ethical judgment, or necessitate nuanced human emotional communication, do I profoundly recognize the limitations of current AI and insist that human professionals must play the leading role, positioning AI merely as an auxiliary tool for information processing or reference?
Commitment to Continuous Learning	□ Do I recognize that AI technology and applications are developing extremely rapidly, and am I consciously, continuously investing time and effort to learn new knowledge, understand new tools, master new skills to ensure my professional competence keeps pace with the times? (Ref Section 9.4)

Checklist 3: Fulfilling the Duty of Diligence - Upholding Responsibility with Intelligent Assistance

Applicable Scenarios: When handling and applying any output generated by AI, whether for internal reference or as part of a final work product (e.g., AI-generated legal research summaries, case lists, risk alert checklists, draft contract clauses, data analysis conclusions).

Consideration Point	Practical Guidance & Self-Check	√/X
Verification & Fact-Checking is Absolutely Core!	□ For any factual information provided by AI (e.g., dates in a case timeline, amounts cited in contract clauses, data from company reports), have I individually cross-checked it against original, reliable sources (original evidence, official documents, authoritative databases)? □ For any legal authority cited by AI (statutes, regulations, case citations), have I personally verified its existence, current validity, accuracy of citation, and applicability to the issue at hand using authoritative legal databases? □ Have I completely and carefully reviewed the entire content generated by AI, not just relying on its provided summary, conclusion, or highlights? □ Have I applied my legal expertise, practice experience, and critical thinking to independently judge the accuracy, logic, completeness, legal reasonableness, and relevance of the AI output to the specific context of the matter? □ Have I been particularly vigilant and thoroughly checked AI outputs that seem “too perfect,” are expressed with unusual fluency and confidence, or significantly contradict my professional intuition or initial assessment (these are often “high-risk zones” for hallucinations)?
Substantive Revision & Professional Value-Add	□ Have I treated any AI-generated draft or suggestion strictly as a “starting point” or “raw material,” and have I made substantial, necessary modifications, additions, deletions, restructuring, and refinements based on it, ensuring the final work product fully meets the specific needs of the case, client objectives, my professional judgment, and the highest quality standards? □ Have I ensured the final presented work product clearly reflects “my” (the human lawyer’s) analysis, judgment, and value, and is not merely a simple copy-paste or minor polishing of the AI output?
Strive for Understanding, Not Blind Obedience	□ For an analytical conclusion or recommendation given by AI, have I made an effort to understand its (possible) reasoning process or basis (e.g., by asking follow-up questions, requesting explanation, or reasoning backwards myself), rather than accepting it without thinking just because “the AI said so”? □ If the AI output conflicts with my initial judgment based on professional experience, have I conducted in-depth, critical analysis to find the specific reasons for the discrepancy (Is the AI wrong? Is my understanding flawed? Or do both have limitations?), rather than easily dismissing my own professional judgment or blindly deferring to the AI’s “authority”?
Allocate Sufficient Review Time	□ In my work planning, have I allocated sufficient, reasonable time for the crucial step of manually reviewing, verifying, and revising AI outputs, rather than mistakenly expecting AI to completely replace my workload, leading to rushed or inadequate review later?
Clear Sense of Final Responsibility	□ Do I deeply and unequivocally understand that regardless of whether and to what extent I used AI assistance in my work, I bear full, non-delegable professional and legal responsibility for all work products (legal opinions, court filings, contracts, research reports, etc.) that I ultimately submit to clients, courts, regulators, or any third party?

Checklist 4: Vigilance Against and Management of Potential Conflicts of Interest - AI Is Not Absolutely Neutral

Applicable Scenarios: When selecting and using AI tools (especially third-party commercial ones), and when using shared AI systems for multiple cases or client matters that might involve conflicts of interest.

Consideration Point	Practical Guidance & Self-Check	√/X
Prudent Vetting of Vendor Background & Potential Biases	□ Before choosing an AI tool (especially for analysis, assessment, or providing recommendations), have I done necessary due diligence on its developer, owner, and major investors? Do they have any background or business connections that could create a potential conflict of interest with my clients, the cases I handle, or my practice area? (e.g., Is the tool developed by a company invested in by my litigation opponent? Is it launched by an industry giant with potentially self-serving biases?) □ Could the AI model’s training data sources or optimization objectives inherently, systematically favor certain positions, viewpoints, solutions, or interest groups? (Often very hard to know fully from the outside, but maintain higher skepticism for tools from vendors with clear stances or commercial interests).
Ensuring Effectiveness of Internal Information Barriers	□ If different teams within my organization (firm/legal dept.) (potentially representing clients or handling matters with direct or indirect conflicts) need to use the same shared AI platform or internal knowledge base, have we established extremely strict, effective technical segregation and access control mechanisms to ensure sensitive information from one matter is absolutely not used by or disclosed to teams working on conflicting matters? (e.g., Is data storage completely segregated? Are access rights strictly need-to-know? Are there robust audit logs?)
Critical Review of Potential Bias in AI Outputs	□ When interpreting and using AI-generated analytical results (risk scores, case recommendations, contract risk ratings, legal point summaries), am I consciously vigilant for potential influences from subtle commercial biases, data biases, or algorithmic design tendencies? Do I proactively cross-reference with other information sources or my own judgment?
Transparent Disclosure When Necessary	□ If, after assessment, a potential connection exists between the AI tool used and the current matter that could potentially impact the objectivity or fairness of the service (even if risk is low), do professional ethics rules require me to make full and timely disclosure to the client or (in appropriate circumstances) the tribunal? (Requires professional judgment based on specific facts and applicable ethics rules).

Checklist 5: Ensuring Transparent Communication & Reasonable Fees - Maintaining Client Trust

Applicable Scenarios: When communicating with clients about the use of AI in service delivery, and when determining fee structures potentially affected by AI application.

Consideration Point	Practical Guidance & Self-Check	√/X
Maintain Appropriate Transparency with Clients on AI Use	□ At the outset of the engagement (e.g., in the engagement letter or initial consultation), or before planning to substantially use AI for key aspects of their matter, have I proactively and clearly explained to the client that we may use AI tools to assist with specific tasks (e.g., enhance research efficiency, accelerate document review, assist drafting standard documents), with the aim of improving service efficiency, controlling costs, or enhancing quality? □ Have I briefly explained the main measures we take to ensure the security and confidentiality of their information when using these tools? □ Have I avoided over-promising or hyping AI’s capabilities to the client, instead objectively explaining its auxiliary role and potential limitations, thereby managing client expectations? □ Have I clearly assured the client that final legal judgment, decisions, and responsibility will always rest with us (human lawyers)?
Reasonableness & Transparency of Fee Structures	□ As AI technology significantly enhances efficiency in certain legal tasks (e.g., drastically reducing hours needed for document review), have I seriously considered and discussed with the client the possibility of adjusting our fee structure? For instance, can we offer more fixed fees, phased fees, or fee arrangements reflecting the actual value created (Value-based Billing), rather than relying solely on traditional hourly billing, which may seem less justified given efficiency gains? □ Regardless of the fee model used, have I provided the client with full, clear, upfront disclosure and explanation of our billing rates, methodology, and (if relevant) how AI usage might impact the fees?

Checklist 6: Upholding Duty of Candor to Tribunal & Evidentiary Responsibilities - Maintaining Judicial Integrity

Applicable Scenarios: When submitting documents or evidence potentially generated with AI assistance to a court or arbitral tribunal, or when citing conclusions based on AI analysis during hearings or arguments.

Consideration Point	Practical Guidance & Self-Check	√/X
Absolute Responsibility for Submissions	□ Have I personally confirmed, 100%, that every word in all written documents I submit to the tribunal (complaints, answers, briefs, evidence lists, etc.), regardless of whether AI assisted in drafting, is truthful, accurate, well-supported, and reflects my independent professional judgment and final approval? □ Have I thoroughly and repeatedly verified all cited cases (names, citations), statutes (numbers, content), and any other facts or data in the documents, ensuring they are real, current, accurately cited, and correctly applied, with absolutely no false information generated by AI “hallucinations”? □ Have I resolutely avoided submitting any core legal arguments or opinions directly generated by AI that I have not substantially reviewed, revised, and ultimately confirmed myself?
Duty to Disclose AI Use (Prudent Judgment)	□ Based on current (possibly unclear) rules in my jurisdiction, specific court requirements, or relevant ethical guidance, do I have an obligation, and to what extent, to disclose to the tribunal and/or opposing counsel that I used artificial intelligence technology in preparing relevant documents or evidence? (e.g., General footnote? Specific labeling of AI-generated parts? Rules are still forming; requires extremely careful judgment based on circumstances, tribunal preferences, risk-benefit analysis. Over-disclosure might invite unnecessary challenges. Consulting senior colleagues or ethics committees may be wise.)
Burden of Proving Reliability of AI-related Evidence	□ If I plan to submit evidence that itself is based on AI analysis or processing (e.g., AI-transcribed audio records, AI-enhanced images/video, an AI data analysis-based risk report), am I fully prepared to clearly, accurately, and credibly explain to the tribunal: · The scientific principles and reliability of the AI technology used? · The specific process, parameters, and methods employed? · The technology’s known accuracy rates, error rates, and limitations?* · Whether the process impacted the authenticity and integrity of the original evidence? □ Have I anticipated that opposing counsel might challenge the admissibility or reliability of this AI-related evidence, and have I prepared corresponding responses? Might I need to retain relevant technical expert witnesses to testify in support of the evidence’s reliability?

Conclusion: Ethical Compass Guides the Way, Professional Responsibility is Paramount

Integrating artificial intelligence into legal practice is far more than a transformation of efficiency and technology; it is a profound ethical practice and adaptation process requiring us to constantly consult our inner ethical compass and scrutinize our actions against the highest professional standards.

As guardians of social rules, fiduciaries of client interests, and staunch defenders of judicial fairness, legal professionals, when wielding the increasingly powerful “intelligent baton” of prompt engineering and AI tools, must constantly keep the string of ethical norms taut. Core professional duties—client confidentiality, competence, diligence, loyalty, transparency, reasonable fees, candor to the tribunal—must be deeply integrated into every interaction with AI and every application of its output.

The checklists above aim to provide a concrete, operational framework for self-examination and risk prevention in daily work. However, final ethical judgments often need to be made in specific, sometimes dilemmatic, complex situations.

Through continuous ethical reflection, prudent decision-making, candid communication, and unwavering adherence to professional bottom lines, we can ensure that artificial intelligence technology truly becomes a responsible, trustworthy force for enhancing legal service quality, improving efficiency, and promoting social fairness and justice, rather than a source of new ethical quandaries, trust crises, or professional risks. The next section will specifically explore the unique risks posed by AI-generated content (AIGC) and strategies for addressing them.