{"id":51571159,"url":"https://github.com/cnaebadi/ai-disclosure-handbook","last_synced_at":"2026-07-10T19:30:59.402Z","repository":{"id":365540112,"uuid":"1272540755","full_name":"cnaebadi/ai-disclosure-handbook","owner":"cnaebadi","description":"A practical guide to AI privacy, profiling, shadow profiling, local AI, cloud AI, and the future of human autonomy.","archived":false,"fork":false,"pushed_at":"2026-06-17T19:28:12.000Z","size":2606,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-17T20:27:46.941Z","etag":null,"topics":["ai","ai-ethics","ai-privacy","ai-safety","artificial-intelligence","cybersecurity","data-privacy","digital-rights","future-of-ai","human-autonomy","llm","local-ai","machine-learning","open-source","privacy","profiling","security","shadow-profiling","surveillance","technology"],"latest_commit_sha":null,"homepage":"","language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/cnaebadi.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-17T17:55:16.000Z","updated_at":"2026-06-17T19:28:16.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/cnaebadi/ai-disclosure-handbook","commit_stats":null,"previous_names":["cnaebadi/ai-disclosure-handbook"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/cnaebadi/ai-disclosure-handbook","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cnaebadi%2Fai-disclosure-handbook","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cnaebadi%2Fai-disclosure-handbook/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cnaebadi%2Fai-disclosure-handbook/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cnaebadi%2Fai-disclosure-handbook/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/cnaebadi","download_url":"https://codeload.github.com/cnaebadi/ai-disclosure-handbook/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cnaebadi%2Fai-disclosure-handbook/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35341768,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-10T02:00:06.465Z","response_time":60,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","ai-ethics","ai-privacy","ai-safety","artificial-intelligence","cybersecurity","data-privacy","digital-rights","future-of-ai","human-autonomy","llm","local-ai","machine-learning","open-source","privacy","profiling","security","shadow-profiling","surveillance","technology"],"created_at":"2026-07-10T19:30:58.606Z","updated_at":"2026-07-10T19:30:59.395Z","avatar_url":"https://github.com/cnaebadi.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# The AI Disclosure Handbook\n\n## What You Should Never Tell an AI — And Why the Real Risk Is Bigger Than You Think\n\n![Cover](assets/cover.jpg)\n\n## Table of Contents\n\n1. [Introduction](#introduction)\n2. [The First Mistake: Thinking Privacy Is About Individual Data](#the-first-mistake-thinking-privacy-is-about-individual-data)\n3. [What You Should Never Share With Any AI](#what-you-should-never-share-with-any-ai)\n4. [Cloud AI vs Local AI](#cloud-ai-vs-local-ai)\n5. [The Puzzle Theory](#the-puzzle-theory)\n6. [Profiling](#profiling)\n7. [Shadow Profiling](#shadow-profiling)\n8. [Why AI Changes Everything](#why-ai-changes-everything)\n9. [Profiling Is Not a Theory](#profiling-is-not-a-theory)\n10. [A Historical Example](#a-historical-example)\n11. [The Difference Between Data and Intelligence](#the-difference-between-data-and-intelligence)\n12. [The New Privacy Problem](#the-new-privacy-problem)\n13. [Why Long Conversations Matter](#why-long-conversations-matter)\n14. [The Profile You Never Intended To Build](#the-profile-you-never-intended-to-build)\n15. [The Hidden Value of AI Conversations](#the-hidden-value-of-ai-conversations)\n16. [The Economics of Behavioral Data](#the-economics-of-behavioral-data)\n17. [What Is Actually Safe To Share?](#what-is-actually-safe-to-share)\n18. [A Better Privacy Question](#a-better-privacy-question)\n19. [Are We Witnessing a Historical Inflection Point?](#are-we-witnessing-a-historical-inflection-point)\n20. [What Is Our Role?](#what-is-our-role)\n21. [Author's Note: A Small Paradox](#authors-note-a-small-paradox)\n22. [Final Thought](#final-thought)\n23. [Discussion](#discussion)\n24. [License](#license)\n\n---\n\n## Introduction\n\nMost AI privacy advice focuses on secrets:\n\n* Don't share passwords.\n* Don't share credit card numbers.\n* Don't upload confidential files.\n\nWhile this advice is correct, it misses a much bigger issue.\n\nThe real danger is often not what you explicitly tell an AI system.\n\nThe real danger is what an AI system can infer from information that appears harmless in isolation.\n\nThis handbook explores:\n\n* AI privacy\n* Cloud AI vs Local AI\n* Profiling\n* Shadow Profiling\n* Behavioral Prediction\n* Human Autonomy\n* The Future Relationship Between Humans and Intelligent Systems\n\n---\n\n## Available Languages\n\n* 🇺🇸 English (Current Document)\n* 🇮🇷 Persian (`/fa/README.md`)\n\n---\n\n# The First Mistake: Thinking Privacy Is About Individual Data\n\nMost people think privacy is about protecting secrets.\n\nPasswords.\n\nBank accounts.\n\nPrivate documents.\n\nAPI keys.\n\nThese are certainly important.\n\nBut modern intelligence systems are often interested in something even more valuable:\n\n**Your profile.**\n\nBecause profiles can be used to predict behavior.\n\nAnd prediction has always been one of the most valuable assets in technology.\n\nThe ability to predict what someone might do tomorrow is often more valuable than knowing what they did yesterday.\n\n---\n\n# What You Should Never Share With Any AI\n\nRegardless of whether you use a cloud AI service or a local model, some categories should never be shared without extreme caution.\n\n## Credentials\n\nNever share:\n\n* Passwords\n* API Keys\n* Authentication Tokens\n* Recovery Codes\n* SSH Keys\n* Private Certificates\n\nA single leak can compromise entire systems.\n\n---\n\n## Customer Data\n\nAvoid sharing:\n\n* Customer records\n* Internal databases\n* User exports\n* Personal information\n\nEven when names are removed, re-identification may still be possible.\n\n---\n\n## Sensitive Information About Other People\n\nAvoid sharing:\n\n* Medical records\n* Legal disputes\n* Private conversations\n* Internal company discussions\n* Confidential negotiations\n\nSomeone else's privacy remains their privacy, even when AI is involved.\n\n---\n\n## Production Infrastructure\n\nNever upload:\n\n* Environment files\n* Production configurations\n* Internal network diagrams\n* Security architecture details\n\nThese assets often contain information far more valuable than source code itself.\n\n---\n\n# Cloud AI vs Local AI\n\nOne of the most common misconceptions is:\n\n\u003e \"If I run AI locally, privacy is solved.\"\n\nReality is more complicated.\n\n---\n\n## Cloud AI\n\nAdvantages:\n\n* Larger models\n* Better performance\n* Faster updates\n* More capabilities\n\nRisks:\n\n* Data leaves your device\n* Third-party infrastructure is involved\n* Future policies may change\n* Users depend on external trust\n\n---\n\n## Local AI\n\nAdvantages:\n\n* Greater control\n* Reduced third-party exposure\n* Better data sovereignty\n\nRisks:\n\n* Device compromise\n* Malware\n* Unauthorized local access\n* Misconfigured systems\n\nLocal AI reduces certain risks.\n\nIt does not eliminate the need for judgment.\n\n---\n\n# The Puzzle Theory\n\nImagine that over several months you tell an AI:\n\n* You use Laravel.\n* You work with PostgreSQL.\n* You own a MacBook.\n* You contribute to open source projects.\n* You are interested in quantitative finance.\n* You are building a security-related Telegram bot.\n\nNone of these statements are secrets.\n\nNone of them identify you directly.\n\nNone appear dangerous.\n\nYet together they create something entirely different:\n\nA profile.\n\nEach statement is a puzzle piece.\n\nThe profile is the completed puzzle.\n\nAnd the completed puzzle often contains information that was never explicitly provided.\n\n---\n\n# Profiling\n\nProfiling is the process of constructing a model of a person using observed behavior and available information.\n\nModern profiling systems may estimate:\n\n* Professional background\n* Interests\n* Purchasing behavior\n* Future intentions\n* Communication style\n* Risk tolerance\n* Decision-making patterns\n\nImportantly:\n\nProfiling does not require certainty.\n\nThe goal is not:\n\n\u003e \"This is definitely true.\"\n\nThe goal is:\n\n\u003e \"This is probably true.\"\n\nFor many commercial systems, probability is enough.\n\n---\n\n# Shadow Profiling\n\nProfiling uses information you knowingly provide.\n\nShadow Profiling goes further.\n\nIt attempts to infer information that you never explicitly disclosed.\n\nExample:\n\nYou never say:\n\n\u003e \"I am planning to move to Germany.\"\n\nInstead, over several months you ask:\n\n* How do German work visas operate?\n* How can I improve my résumé?\n* How do German taxes work?\n* How does German healthcare work?\n* What is the cost of living in Berlin?\n\nNo individual question reveals your plan.\n\nTogether they may reveal it quite clearly.\n\nThe conclusion was never stated.\n\nIt emerged.\n\nThat is Shadow Profiling.\n\n---\n\n# Why AI Changes Everything\n\nProfiling existed long before AI.\n\nAdvertising companies have spent decades building behavioral profiles.\n\nRecommendation systems have spent decades predicting preferences.\n\nSocial networks have spent decades analyzing engagement.\n\nWhat changes with AI is scale.\n\nHumans struggle to connect thousands of weak signals.\n\nMachines do not.\n\nHumans forget conversations from six months ago.\n\nMachines can analyze them instantly.\n\nHumans miss subtle correlations.\n\nMachines are designed to find them.\n\nThe result is a world where seemingly harmless information becomes increasingly valuable when aggregated.\n\n---\n\n# Profiling Is Not a Theory\n\nWhen people hear the word \"profiling,\" they often imagine a futuristic technology that belongs in science fiction.\n\nIn reality, profiling has been part of the digital economy for decades.\n\nLong before modern AI systems existed, companies were already collecting signals from:\n\n* Search queries\n* Website visits\n* Purchase histories\n* Click patterns\n* Device information\n* Location data\n\nThe objective was simple:\n\nBuild increasingly accurate models of human behavior.\n\nThe emergence of AI did not create profiling.\n\nIt increased the speed, scale, and sophistication of profiling.\n\n---\n\n## A Historical Example\n\nOne of the most frequently cited examples in discussions about predictive analytics involved retail purchasing behavior.\n\nBy analyzing shopping patterns, data scientists discovered that seemingly unrelated purchases could sometimes predict major life events before customers explicitly announced them.\n\nThe lesson was not that companies could read minds.\n\nThe lesson was that patterns often reveal more than individual facts.\n\nThis principle applies far beyond retail.\n\nThe same logic can be applied to careers, interests, habits, relationships, and future intentions.\n\n---\n\n# The Difference Between Data and Intelligence\n\nA common misconception is that data itself is valuable.\n\nData is rarely the final product.\n\nThe real value often comes from transforming data into predictions.\n\nConsider the difference:\n\nData:\n\n* A person searched for apartment prices.\n* A person searched for visa requirements.\n* A person searched for taxation rules.\n\nIntelligence:\n\n* This person may be preparing to relocate internationally.\n\nThe individual facts are not particularly useful.\n\nThe inferred conclusion is.\n\nThis distinction becomes increasingly important in the age of AI.\n\n---\n\n# The New Privacy Problem\n\nHistorically, privacy discussions focused on collection.\n\nWho collected data?\n\nHow much data was collected?\n\nWhere was it stored?\n\nThose questions remain important.\n\nHowever, AI introduces an additional layer:\n\nInference.\n\nThe challenge is no longer limited to protecting information.\n\nThe challenge increasingly involves protecting the conclusions that can be generated from information.\n\nThis creates a difficult question.\n\nCan a person meaningfully protect their privacy if the most sensitive information about them is never explicitly stated, but instead inferred?\n\n---\n\n# Why Long Conversations Matter\n\nTraditional search engines typically receive short requests.\n\nAI systems increasingly receive context-rich conversations.\n\nPeople explain situations.\n\nThey describe emotions.\n\nThey provide background information.\n\nThey discuss future plans.\n\nThe result is not merely more data.\n\nThe result is higher-quality signals.\n\nA thousand isolated search queries may reveal less about a person than a single six-month conversation history.\n\nThis is one reason why conversational AI deserves a different privacy discussion than traditional search.\n\n---\n\n# The Profile You Never Intended To Build\n\nMost users do not consciously build a profile.\n\nIt emerges naturally.\n\nA question about taxes.\n\nA question about relationships.\n\nA question about careers.\n\nA question about health.\n\nA question about finances.\n\nEach appears insignificant.\n\nCollectively they may become one of the most detailed portraits a person has ever created of themselves.\n\nSometimes more detailed than the profile they would provide to a friend.\n\nSometimes more detailed than the profile they would provide to an employer.\n\nSometimes more detailed than the profile they would consciously write themselves.\n\nThat does not automatically imply danger.\n\nBut it does imply responsibility.\n\nBecause every powerful model begins with understanding.\n\nAnd every profile is ultimately an attempt to understand.\n\n---\n\n# The Hidden Value of AI Conversations\n\nMany people interact with AI differently than they interact with search engines.\n\nThey discuss:\n\n* Career plans\n* Business ideas\n* Financial concerns\n* Personal fears\n* Creative ambitions\n* Relationship problems\n\nFor the first time in history, millions of people are voluntarily engaging in long-form conversations with systems capable of analyzing those conversations.\n\nWhether those conversations remain private, how they are governed, and how future systems may use them are among the most important questions of our era.\n\n---\n\n# The Economics of Behavioral Data\n\nWhy does this matter?\n\nBecause behavior is valuable.\n\nPrediction is valuable.\n\nAttention is valuable.\n\nHuman decisions are valuable.\n\nHistorically, some of the world's largest technology companies built their businesses around understanding and predicting human behavior.\n\nAI has the potential to dramatically accelerate these capabilities.\n\nThis does not automatically imply abuse.\n\nNor does it automatically imply safety.\n\nIt simply means the incentives are significant.\n\nAnd significant incentives deserve scrutiny.\n\n---\n\n# What Is Actually Safe To Share?\n\nThere is no universal answer.\n\nBut some categories of information are generally lower risk than sensitive, identifying, or behaviorally revealing data.\n\nExamples that are usually safer include:\n\n* General educational questions\n* Concept explanations\n* Rewriting non-confidential text\n* Generating fictional examples\n* Learning new skills\n* Analyzing public or synthetic data\n\nEven then, context matters.\n\nA single general question may reveal very little.\n\nA long sequence of general questions can still form a pattern.\n\nSo the issue is not only the type of data.\n\nIt is also the overall pattern of interaction.\n\n---\n\n# A Better Privacy Question\n\nMost people ask:\n\n\u003e \"What should I not tell AI?\"\n\nA better question may be:\n\n\u003e \"What can AI infer from everything I tell it?\"\n\nThese are not the same question.\n\nAnd the second question is often more important than the first.\n\n---\n\n# Are We Witnessing a Historical Inflection Point?\n\nThis article is not an argument against artificial intelligence.\n\nAI has already helped people learn faster.\n\nBuild faster.\n\nCreate faster.\n\nSolve problems faster.\n\nIts benefits are undeniable.\n\nHowever, every transformative technology changes power structures.\n\nThe printing press changed information.\n\nThe internet changed communication.\n\nArtificial intelligence may change observability.\n\nFor the first time, it is becoming technically feasible to build systems capable of continuously modeling human behavior at unprecedented scale.\n\nWhether these capabilities are ultimately used for empowerment, optimization, surveillance, influence, or control remains one of the defining questions of our generation.\n\nThe future is not predetermined.\n\nBut neither is it guaranteed.\n\n---\n\n# What Is Our Role?\n\nPerhaps the most important question is not:\n\n\u003e \"What should I tell an AI?\"\n\nPerhaps the real question is:\n\n\u003e \"What kind of relationship should humans have with systems that can learn so much about them?\"\n\nThe first responsibility is understanding.\n\nThe first step in solving any problem is recognizing that it exists.\n\nAnd if we do not yet know exactly what actions should be taken, there is still something meaningful we can do.\n\nWe can discuss these concerns.\n\nWe can challenge assumptions.\n\nWe can ask difficult questions.\n\nWe can help others understand the trade-offs.\n\nMany of the most important changes in history did not begin with solutions.\n\nThey began with awareness.\n\nA society that understands a problem is far more capable of solving it than a society that ignores it.\n\nIf enough people develop a shared understanding of these challenges, the safeguards, technologies, policies, and ideas required to address them are more likely to emerge.\n\nSometimes the first spark of meaningful change is not a solution.\n\nSometimes it is simply a conversation.\n\n---\n\n# Author's Note: A Small Paradox\n\nLife is full of paradoxes.\n\nThe initial idea for this article was sparked during a conversation with an artificial intelligence system.\n\nThe decision to pursue the idea, challenge it, and shape its direction came from a human.\n\nParts of the editing, structuring, refinement, and translation were assisted by artificial intelligence.\n\nAt one point, the AI itself resisted incorporating some of the arguments presented in the final section, suggesting caution against fear-mongering, exaggeration, or unsupported conclusions.\n\nA human insisted on keeping the discussion alive.\n\nThe final result became a negotiation.\n\nNot purely human.\n\nNot purely artificial.\n\nA collaborative product of both.\n\nAnd perhaps that is fitting.\n\nAfter all, this article is ultimately about the relationship between humans and intelligent systems.\n\nOne final detail:\n\nWe intentionally chose not to mention which AI system participated in the process.\n\nNot because it is secret.\n\nBut because we would prefer readers to engage with the ideas rather than immediately taking sides based on the name attached to them.\n\nSometimes labels attract more attention than arguments.\n\nAnd sometimes the argument is the part that matters most.\n\n---\n\n# Final Thought\n\nPrivacy is no longer only about hiding information.\n\nPrivacy is increasingly about controlling the patterns that emerge from information.\n\nAnd in the age of artificial intelligence, that difference may matter more than most people realize.\n\n---\n\n## Discussion\n\nQuestions, criticism, corrections, and alternative viewpoints are welcome.\n\nThis repository is intended as a starting point for discussion rather than a definitive answer.\n\n---\n\n## License\n\nThis work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).\n\nhttps://creativecommons.org/licenses/by/4.0/\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcnaebadi%2Fai-disclosure-handbook","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fcnaebadi%2Fai-disclosure-handbook","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcnaebadi%2Fai-disclosure-handbook/lists"}