Privacy, Disclosure & Liability

The most common and avoidable AI risk in daily practice is quietly leaking confidential information by pasting it into a public tool. When you enter text or images into a free consumer AI service, that data leaves your control: depending on the service and its settings, it may be stored on the vendor's servers, reviewed by staff, and - critically - used to train future models. Content used for training can, in principle, resurface in outputs to other users. Treat anything you paste into a public tool as potentially no longer private.

So draw a hard line around what must never go into a public AI service without proper safeguards: client personal data (names, contacts, anything identifying a person - squarely covered by privacy laws like the GDPR, India's DPDP Act, and others); confidential project information under an NDA; unpublished designs and drawings, your own or a client's; security-sensitive details (site security, structural vulnerabilities); and financial or contractual specifics. The safe workflow is to anonymise and generalise before you prompt - strip names, addresses, and identifying specifics, and ask the question in the abstract ('a 4-storey office in a hot-humid climate' rather than the real client, address, and drawings). For genuinely sensitive work, use tools with proper data agreements: enterprise or business tiers that contractually promise not to train on your data, or self-hosted and private-deployment models. Read the data-handling terms of every tool the way the last lesson had you read the IP terms - before you paste, not after.

Two forces make this more than good manners. The first is data-protection law. Regimes like the EU's GDPR - and their growing counterparts worldwide, including India's Digital Personal Data Protection Act - impose real obligations on how you handle personal data: a lawful basis for processing, limits on transferring it (including to AI vendors, potentially across borders), and serious penalties for breaches. Feeding a client's personal data into a consumer AI tool can constitute an unlawful disclosure or transfer, and 'I did not realise the tool stored it' is not a defence. Exactly how these laws apply to your situation is a question for a data-protection professional; the point here is to recognise that the obligation exists and to build your workflow to respect it.

The second force is your own professional duty of confidentiality. Architects, designers, and their firms hold client information in confidence as a matter of professional conduct and, usually, contract - NDAs and appointment agreements routinely bind you to protect it. Pasting confidential project material into a public tool can breach that duty regardless of whether any law is triggered, exposing you to a claim and, worse, a loss of trust. The safe posture is to treat AI tools as third parties to whom you are disclosing information - because that is what they are - and to disclose only what you would be comfortable, and permitted, to share with any outside party. When in doubt, anonymise, or do not paste it at all.

Alongside protecting client data sits a duty of candour: being appropriately transparent that you used AI. The norm is shifting quickly toward disclosure. Clients increasingly want to know when AI materially shaped their deliverables; some contracts, competitions, and public-sector briefs now require it; and professional bodies are issuing guidance. Beyond any rule, honesty is simply the trust-preserving choice - a client who later discovers undisclosed AI use may feel deceived, which is far more damaging than the disclosure would ever have been.

Disclosure does not mean annotating every keystroke, and the right level is proportionate to the reliance and the context. You would not flag using a spell-checker; you would flag that concept imagery was AI-generated, that a report draft was AI-assisted, or that an analysis leaned on an AI model - anything a reasonable client would want to know in judging the work. The practical approach: agree the use of AI with your client up front, ideally in the appointment or a short AI-use clause your lawyer helps word (this ties directly to the IP and confidentiality terms from Lesson 9.3); set expectations about where AI is and is not used; and keep it a normal, unembarrassed part of how you describe your process. Transparency also protects you - a client who knew and agreed to your AI-assisted method is a client who cannot later claim they were misled.

Here is the hard core of the whole module. When AI output is wrong and you used it in your professional work, the liability is yours - not the tool's. If an AI-summarised code clause was incorrect and your building fails an inspection; if an AI-drafted spec contained an error that reached site; if an AI analysis was wrong and a decision was made on it - you, as the professional who adopted and issued that work, carry the responsibility. The AI vendor's terms almost universally disclaim liability for output, and a court is exceedingly unlikely to accept 'the AI told me so' as a defence for professional negligence. Using a tool never transfers your duty of care.

This is exactly why the human-in-the-loop is not merely good practice but professional self-protection, and why Lesson 9.1's rule - scrutiny scales with stakes - is really a liability-management rule. Every high-stakes AI output you issue must be verified against primary sources as if you had produced it yourself, because in the eyes of your duty of care, you did. Two further safeguards matter: check whether your professional indemnity insurance covers AI-assisted work (policies are evolving, and you do not want to discover a gap after a claim - ask your insurer), and keep a record of your verification on significant work, so you can show the human judgement you applied. Far from making you replaceable, this responsibility is precisely why skilled professionals remain essential: someone has to own the outcome, and that someone is you.

Notice how neatly this closes the argument the whole course has made. The reason AI does not replace the designer is the same reason the liability cannot move to the tool: professional practice is not the production of drawings and documents - a machine can help make those - it is the accountable exercise of judgement on someone else's behalf. Accountability is the thing that cannot be automated, because it is fundamentally a human, legal, and relational commitment, not an information-processing task. So the liability staying with you is not an unfair burden bolted onto a helpful technology; it is the very definition of what your client is retaining you for. Embrace that, and AI stops looking like a threat to your role and starts looking like what it is - leverage applied under your responsibility.

These duties combine into a simple protocol you can run as habit. Before you prompt: ask whether this content is safe to share with an outside party; if it contains personal, confidential, or unpublished material, anonymise it or switch to a no-train, data-protected tool. While you work: keep AI in a supporting role, ground it in real sources, and evaluate every output with scrutiny scaled to the stakes. Before you issue: verify anything consequential against primary sources, and make sure your human judgement - not the model's confidence - is what stands behind the deliverable. With your client: disclose material AI use, ideally agreed in advance, and be ready to explain how you checked the work. Behind the scenes: confirm your insurance, tool terms, and contracts cover how you actually work.

Run that protocol and you have closed the loop the entire course has been building - direct, generate, evaluate, refine - not just for quality but for privacy, honesty, and accountability. This is what separates AI-assisted professional practice from careless AI use: the same tools, but wrapped in the judgement, discretion, and responsibility that a machine cannot supply and a client is paying you for. The final module turns from the individual to the studio - how to adopt AI across a team and a career - but the foundation is here: use AI freely, stay the author, protect your clients, be honest, and own the result.