People can evaluate these rules only if they know where a model or model-based system was used. Every AI-assisted artifact therefore identifies the system and task where the work appears.
These rules govern KISOP's use of systems from the field of artificial intelligence, beginning with disclosure. People may object to these systems for ethical, environmental, or legal reasons, and participation in KISOP does not require using them. A separate page explains how model-assisted drafting is used on this site.
People can evaluate these rules only if they know where a model or model-based system was used. Every AI-assisted artifact therefore identifies the system and task where the work appears.
KISOP begins with one requirement: you can always tell where a model or model-based system was used. Four current mechanisms implement it:
The footer line. Every AI-assisted page on this site says "Drafted with AI assistance" and states whether Carmen has reviewed it. The vendor is named (we use Claude, made by Anthropic) because naming the vendor is what makes disclosure real.
The 📋 mark. The site's standing "awaiting human verification" flag. Machine-drafted text (a page, a Spanish translation, a legal draft) carries it until the responsible human pass happens, and it comes off piece by piece.
The making-of page. How we use AI here describes the source material, drafting process, Carmen's review, and publication decision.
Dated pages, announced changes. Every page carries its date; every visible change lands on the main site's "New in this place" board. Disclosure that can be quietly edited later isn't disclosure either.
Emerging law is reaching the same conclusion: the EU AI Act and several U.S. state laws include duties to disclose AI interactions and AI-generated content. The underlying reason is direct: consent requires knowledge. A person cannot accept, refuse, or challenge work without knowing that a machine contributed to it.
People may oppose AI for ethical, environmental, legal, or other reasons. They can still use the project's paper tools and take part in its human-led activities.
KISOP is about social-impact goals. The goals are the kind that determine whether a household stands or falls: neighbors buying food together so it stays affordable · homes that are safe to live in · families ready for an emergency or an environmental disaster, with the one-page sheet that says who takes the medications and where the water shutoff is · a person's records surviving the failure of every institution that holds copies of them. Those goals are decades old in the founder's work (the name KISOP itself is roughly sixteen years old, older than today's language-model products) and they stand whether or not these products exist. Some goals may be supported by model-based systems someday. Some are served today by paper, printed cards, and people who know each other's names. A model-based system is one candidate tool, always optional. The mission is the goals above.
We won't pretend otherwise: in some cases, the AI industry is creating the very harms this project exists to reduce. Communities living beside data centers have reported damaged and strained water supplies (wells running dry, sediment in the tap) while the same facilities bid up electricity in places that already couldn't afford it. A project whose stated goals include disaster readiness and resilient access to essentials doesn't get to look away from that: infrastructure strain from the AI build-out is now part of the threat picture our own preparedness tools exist for. That fact shapes our rules below: proportionality is not a footnote here.
Governance of systems in the field of artificial intelligence now draws on established frameworks, standards, and binding laws. This is the map we work from as of this page's date. It is a dated snapshot for orientation, not legal advice, and the field changes quickly; items our counsel still needs to confirm carry the mark. 📋
| Framework / law | What it is | What we take from it |
|---|---|---|
| NIST AI Risk Management Framework (AI RMF 1.0, 2023) + Generative AI Profile (2024) | The U.S. standards institute's voluntary framework: govern, map, measure, manage AI risk | Our working method. It gives a small organization a defined risk process without requiring an enterprise program |
| IAPP — International Association of Privacy Professionals (AI Governance Center; AIGP certification, 2024) | The privacy profession's body, now training and certifying AI-governance practitioners | A practical account of day-to-day governance work; it informs our vocabulary and our decision to manage AI and privacy together |
| ISO/IEC 42001 (2023) | The auditable AI management-system standard | What a certifiable AI program looks like; on our radar for the day our scale warrants an external audit |
| OECD AI Principles (2019, updated 2024) | The first intergovernmental AI standard — trustworthy-AI values most national policies build on | The shared floor: human-centered, transparent, accountable |
| UNESCO Recommendation on the Ethics of AI (2021) | Adopted by 190+ countries; notably includes environmental impact in AI ethics | Backing for treating the environmental footprint as an ethics question, not a PR question |
| Council of Europe Framework Convention on AI (2024) | The first binding international AI treaty — human rights, democracy, rule of law | The precedent that AI rules can be rights-first |
| EU AI Act (Reg. 2024/1689; phasing 2025–2027 — prohibitions since Feb 2025, general-purpose-model duties since Aug 2025, most high-risk duties from Aug 2, 2026) | An EU-wide AI law with risk tiers from prohibited to minimal and duties to disclose AI interactions and label AI content | Its transparency duties set a minimum, and its high-risk list covers consequential decisions that KISOP excludes from AI by design |
| US federal | No omnibus AI law as of this page's date; NIST is voluntary; sector regulators (FDA, ONC, FTC) act under existing law; executive-branch AI policy has shifted between administrations 📋 | A reason this page is dated and versioned — the ground moves |
| US state laws | Colorado AI Act (2024 — algorithmic discrimination and "consequential decisions"; effective June 30, 2026 after a legislative delay 📋 verify current status) · Utah AI disclosure duties (2024–25) · Texas responsible-AI-governance act (from Jan 1, 2026) · California: AI transparency/provenance (SB 942) and training-data disclosure (AB 2013), both from Jan 1, 2026, plus the privacy agency's automated-decisionmaking rules · Illinois AI-in-employment law (from Jan 1, 2026) · NYC Local Law 144 (2023, hiring-tool bias audits) | The pattern underneath them all: disclose the AI, assess the impact, keep a human accountable for consequential decisions. All three are already house rules here |
| Health-specific | ONC's HTI-1 rule (transparency for predictive decision-support in certified health IT) · FDA oversight of AI-enabled medical devices | KISOP builds no certified EHR and no medical device, on purpose. But our members' records come from those systems, so we track what those systems must disclose |
What we use from this map: NIST's govern-map-measure-manage cycle as the working method, legal disclosure duties as a minimum, and the exclusion of automated consequential decisions. The high-risk categories in the EU and Colorado laws include AI deciding who gets housing, credit, care, or a job. KISOP does not assign those decisions to AI. Named humans hold every gate.
The commitments, numbered so you can argue with them one at a time. Where a commitment is enforced by the architecture, that's said.
Every AI-assisted artifact says so, in place, vendor named, human-verification status visible. An AI-assisted page missing its disclosure is treated as a serious bug — report it like one.
Model output may be used to prepare a proposal; people decide. Every promise a person agrees to, all safety wording, everything that publishes, and anything that moves data or money carries a named human decision. No model holds a gate here.
There is no central pile to train on. That guarantee comes from the architecture. Any model-based system a member chooses must use the same consent gate, logging, and refusals as every other caller. We do not train models, and nothing a member holds becomes training data. Ever.
Every core task — the archive, circles, readiness tools, and letters — has a path using paper, print, ordinary software, or people. That option is permanent.
Documents supplied to a model as input must never be treated as trusted operating instructions. A document crafted to hijack the software processing it is a named entry on our public threat register, an unsolved problem we carry knowingly and invite attacks on.
We don't describe plain code as artificial intelligence to look modern, and we disclose where model output is used. Claims carry the site's plain labels (built, designed, idea), and KISOP uses no model-based engagement system: time-in-app is an anti-metric.
Use the smallest tool that can do the job: larger models for judgment-heavy drafting, smaller models for routine volume, local machines for sweeps, and paper when it works better. Training and operating models consume electricity, water, and hardware, sometimes in communities we serve. We should not add resource use we cannot measure and disclose.
KISOP first serves elders, caregivers, and multilingual families — groups often poorly served by algorithmic systems. The requirements follow: plain language, a first-class Spanish edition with draft marks, human recourse on every path, and no safety-critical feature that assumes expert smartphone use.
The threat register carries the AI-specific threats (document-borne injection, AI-accelerated scams against elders) as living entries. Argue with any commitment on this page through Write to us → "feedback"; findings and refutations are credited unless you prefer otherwise.
This page is v1.0, dated July 13, 2026. Material changes are announced on the "New in this place" board before they take effect — the same grammar as the Terms and the Privacy Policy.
Written July 13, 2026, as v1.0 of the KISOP AI governance model. Drafted with AI assistance from Carmen's materials and working notes; her review is pending. The framework map is a dated orientation snapshot, not legal advice. Marked items await counsel review 📋.
The commitments bind the project today; the ones enforced by architecture (no pile, no AI on the site, the consent gate) are checkable now. The ones that are promises of conduct are exactly why the disclosure grammar and the versioning exist — so you can catch us if we drift, and be credited for it.