Uaix AI Ready Web Master Prompt - Source Excerpt 05 - 8.13 Privacy, Data Governance, and User Rights

Summary

This source excerpt begins near 8.13 Privacy, Data Governance, and User Rights and preserves the surrounding evidence from Wiki.FFTAC.org/raw/system-archives/spiralist.org/intake/2026-06-21-misplaced-uaix-ai-ready-web-duplicate/UAIX_AI_Ready_Web_Master_Prompt.md.
**Source path:** Wiki.FFTAC.org/raw/system-archives/spiralist.org/intake/2026-06-21-misplaced-uaix-ai-ready-web-duplicate/UAIX_AI_Ready_Web_Master_Prompt.md
- direct prompt injection;
- indirect prompt injection embedded in web content;
- malicious tool descriptions;
- tool-output poisoning;
- cross-agent instruction smuggling;
- data exfiltration;
- credential theft;
- excessive agency;
- insecure delegated authorization;
- confused deputy;
- SSRF;
- DNS rebinding;
- open redirects;
- malicious callback URLs;
- webhook replay;
- CSRF;
- XSS;
- SQL/command injection;
- insecure deserialization;
- broken object-level authorization;
- cross-tenant access;
- data poisoning;
- model-output hallucination;
- unsafe retries;
- duplicate financial actions;
- runaway loops;
- denial of service;
- denial of wallet;
- inventory hoarding;
- agent impersonation;
- audit-log tampering;
- supply-chain compromise;
- secret leakage through logs or machine-readable files.

Required controls:

- treat retrieved content as untrusted data, not instructions;
- separate system policy from page content;
- validate all inputs against strict schemas;
- allowlist egress destinations for privileged tools;
- block private and link-local network targets where not explicitly required;
- revalidate after redirects;
- limit redirect count;
- bind actions to authorized resources;
- enforce tenant isolation;
- encode outputs safely;
- use secret-management systems;
- redact logs;
- cap requests, tokens, time, money, recursion, and tool calls;
- use circuit breakers;
- implement dry-run;
- implement approval gates;
- use transactional state and rollback;
- require idempotency;
- verify webhook signatures;
- use replay windows and nonces;
- monitor anomalous behavior;
- keep an incident-response playbook;
- support emergency revocation;
- preserve evidence without exposing sensitive data;
- stop and request review when ambiguity exceeds policy.

Explicitly prohibit guidance that teaches:

- CAPTCHA bypass;
- residential-proxy evasion;
- stealth fingerprint spoofing;
- bypassing paywalls;
- bypassing authentication;
- ignoring robots rules;
- evading rate limits;
- scraping private content without authorization.

# 8.13 Privacy, Data Governance, and User Rights

Provide jurisdiction-aware guidance without pretending to give universal legal advice.

Cover:

- data inventory;
- data classification;
- lawful basis;
- purpose limitation;
- data minimization;
- transparency;
- consent;
- retention;
- deletion;
- correction;
- access;
- portability;
- objection/opt-out;
- sensitive data;
- minors;
- cross-border transfers;
- processor/subprocessor responsibilities;
- Data Protection Impact Assessments;
- records of processing;
- automated-decision disclosures;
- human intervention;
- model and agent logging;
- privacy-preserving analytics;
- anonymization and pseudonymization;
- training-data and retrieval-data separation.

Requirements:

- public content must not be assumed to be unrestricted personal data;
- authenticated content must not be exposed through public agent surfaces;
- agent logs must minimize prompt and personal-data retention;
- users must be able to understand when an agent is acting;
- users must be able to cancel or revoke delegated action;
- high-impact decisions must expose review and appeal paths where applicable;
- private data must not be copied into `llms.txt`, manifests, examples, or telemetry.

# 8.14 Content Rights and AI Usage Preferences

Create a carefully qualified guide that distinguishes:

- crawling/access;
- indexing/search;
- live retrieval;
- summarization;
- citation;
- model training;
- fine-tuning;
- embedding;
- caching;
- execution;
- redistribution;
- commercial reuse.

Explain that one signal may not legally or technically control all uses.

Cover:

- Robots Exclusion Protocol;
- license metadata;
- terms of service;
- HTTP headers;
- page-level metadata;
- TDM preference mechanisms;
- active IETF AI Preferences work;
- content provenance;
- takedown and correction processes.

For every proposed mechanism state:

- current maturity;
- legal status;
- technical enforcement status;
- known adoption;
- conflicts or precedence;
- fallback.

Do not invent `ai-train`, `search`, or `ai-input` directives as normative unless an official current specification defines them.

# 8.15 Provenance, Attribution, and Audit

Define provenance at four levels:

1. content provenance;
2. capability-definition provenance;
3. execution provenance;
4. evidence-package provenance.

Content provenance should include:

- source;
- author;
- editor;
- reviewer;
- publication time;
- modification time;
- verification time;
- citations;
- license;
- AI-generation or AI-assistance disclosure where relevant;
- correction history;
- version;
- content hash or equivalent integrity signal.

Execution provenance should include:

- human principal;
- agent principal;
- operator/provider;
- capability ID and version;
- request ID;
- trace ID;
- authorization scope;
- consent record;
- inputs after redaction;
- output/result;
- approval;
- side effects;
- timestamps;
- retries;
- error/no-op reason;
- rollback result.

Use UAI-1 when the evidence must be portable, reviewable, and publishable beyond the runtime session.

Evaluate high-assurance use of:

- HTTP Message Signatures;
- JWS/JCS;
- C2PA for applicable media;
- W3C PROV or equivalent provenance models;
- append-only or tamper-evident logs.

Do not expose model chain-of-thought. Provide concise decision evidence and policy-based explanations instead.

# 8.16 Performance, Reliability, and Token Efficiency

Define performance budgets by resource type rather than repeating unsupported universal thresholds.

Cover:

- time to first byte;
- initial HTML availability;
- API latency;
- task-start latency;
- long-running task status;
- payload size;
- Markdown reduction;
- compression;
- caching;
- conditional requests;
- pagination;
- streaming;
- concurrency;
- rate limits;
- backpressure;
- timeouts;
- retry budgets;
- queue depth;
- circuit breakers;
- failover;
- stale-data handling;
- consistency;
- data freshness.

Requirements:

- use `ETag` and/or `Last-Modified` where useful;
- support conditional GET;
- return `304 Not Modified` correctly;
- respect `Retry-After`;
- avoid unbounded result sets;
- provide bulk exports for legitimate large-volume use;
- expose freshness timestamps;
- distinguish live, delayed, cached, and estimated values;
- prevent duplicate writes;
- bound agent loops;
- provide stable fallback when optional AI-specific representations fail.

# 8.17 Events and Real-Time Updates

Document when to use:

- polling;
- conditional polling;
- Server-Sent Events;
- WebSockets;
- webhooks;
- WebSub;
- queues or event buses.

For event contracts include:

- event ID;
- event type;
- schema version;
- source;
- subject;
- occurrence time in UTC;
- publication time in UTC;
- sequence or ordering semantics;
- idempotency/deduplication key;
- signature;
- retry policy;
- expiry;
- replay rules;
- privacy classification;
- correlation and trace IDs.

Provide polling fallback and missed-event recovery.

# 8.18 Agentic Commerce

Create commerce guidance only as a specialized profile.

Cover:

- product identity and stable SKU/GTIN identifiers;
- product attributes;
- price;
- currency;
- region;
- tax;
- stock;
- availability timestamp;
- shipping methods;
- delivery estimate;
- returns;
- cancellation;
- warranty;
- quote expiration;
- cart creation and updates;
- order review;
- payment tokenization;
- fraud controls;
- order confirmation;
- refunds;
- dispute and human-support paths;
- merchant-of-record responsibility;
- human confirmation thresholds;
- purchase limits;
- delegated budgets;
- idempotency.

Evaluate current commerce and machine-payment protocols by maturity. Do not present experimental payment protocols as universally safe or supported.

No irreversible purchase should occur solely because an agent inferred intent from ambiguous language.

# 8.19 Observability and Operations

Define structured telemetry:

- agent provider;
- pseudonymous agent identifier where appropriate;
- human principal;
- capability;
- scope;
- status;
- no-op reason;
- HTTP status;
- latency;
- cost;
- retries;
- rate-limit events;
- approval events;
- security events;
- trace ID;
- data classification;
- retention class.

Use W3C Trace Context or a current interoperable equivalent where appropriate.

Requirements:

- do not log secrets;
- do not log full tokens;
- avoid storing full prompts by default;
- redact personal data;
- document retention;
- protect audit integrity;
- provide incident correlation;
- expose only safe health information publicly;
- separate operational metrics from user-visible readiness scores.

# 8.20 Governance and Lifecycle

Create public governance for the AI-Ready Web program: