Here is what my product does, where my users are, and where my infrastructure runs.<br/><br/>List every AI-specific regulation or proposed regulation that plausibly applies, and for each state:<br/>1. Whether it is in force, proposed, or speculative<br/>2. The specific provision that reaches my case<br/>3. What it requires me to do<br/><br/>Separate in-force obligations from proposals clearly. Do not blend them into one list - I need to know what I must do now versus what I should watch.<br/><br/>If a regulation probably does not apply to me, say so and say why.<br/><br/>DEPLOYMENT:<br/>[paste]
A draft bill covers AI that "exceeds human cognitive performance and capabilities across most domains." There is no compute threshold, benchmark, or evaluation standard attached.<br/><br/>Take my system described below and argue BOTH sides:<br/>1. The case a regulator would make that it falls inside this definition<br/>2. The case I would make that it falls outside<br/><br/>Then tell me which specific facts about my system decide it, and which of those facts I could document now to make the second argument easier later.<br/><br/>SYSTEM:<br/>[paste]
Given the obligations below that apply to my deployment, audit what I would need to produce if a regulator asked tomorrow.<br/><br/>For each obligation: what evidence satisfies it, whether I appear to have that evidence based on what I have described, and what I would have to reconstruct.<br/><br/>Rank the gaps by how hard they are to fill retroactively. Anything that requires contemporaneous records - training data provenance, evaluation results at time of release, incident logs - goes at the top, because those cannot be recreated after the fact.<br/><br/>OBLIGATIONS AND CURRENT STATE:<br/>[paste]
My product depends on the model providers listed below.<br/><br/>For each, identify regulatory events that would disrupt my service, not theirs: export controls on their models, a development pause reaching their release cycle, jurisdiction restrictions, or a forced change to their terms.<br/><br/>For each risk, tell me what my fallback would be and how long switching would take. Be specific about whether my prompts, fine-tunes and evaluation suites would transfer.<br/><br/>PROVIDERS AND DEPENDENCIES:<br/>[paste]
Draft a one-page statement of how my system works, what it can and cannot do, and what controls are in place.<br/><br/>Write it for a non-technical regulator or journalist, not an engineer. Requirements:<br/>1. No claims I cannot evidence<br/>2. State limitations plainly rather than burying them<br/>3. Name the specific controls, not "robust safeguards"<br/>4. Flag in brackets any claim where I need to check a fact before publishing<br/><br/>SYSTEM AND CONTROLS:<br/>[paste]
Below are compliance claims and requirements a vendor has told me apply to my use of their product.<br/><br/>For each, tell me whether it is: an actual legal obligation, a contractual obligation they have imposed, an industry norm with no force, or marketing.<br/><br/>Cite the source of the obligation where one exists. Where a vendor is presenting their own product requirement as a regulatory one, say so plainly.<br/><br/>VENDOR CLAIMS:<br/>[paste]