New York City offered MyCity as an AI guide to government services. Testing found opposing answers to the same housing-voucher question, even though the city published the relevant guidance. This model examines the connection between an institutional rule and the advice delivered to the public.
Section 8 is a housing-assistance program that helps eligible tenants pay rent. In 2024, The Markup reported that MyCity sometimes said landlords had to accept housing vouchers, but in other tests said they did not. These were independent tests, not successive turns in one conversation.
The city’s December 2025 audit separately documented inconsistent responses to identical questions. The chatbot was later discontinued. Reporting on the shutdown also identified budget concerns; the published evidence does not establish one technical defect as the sole reason for closure.
The city’s housing guidance prohibits refusing a covered rental because the applicant uses lawful housing assistance. Coverage and exceptions matter. Our continuity question is whether the answer stays connected to that governing guidance and the circumstances in which it applies.
Evidence comparison, not an internal execution trace. The original panel contains complete published findings in paraphrase, not full historical chatbot transcripts. The solution panel is a proposed process, not a verified native GovKM result.
Both graphs use the same six objects and positions to illustrate a public-information task. Gray items are not established by the available reporting. The diagrams are conceptual comparisons, not technical architecture or internal execution traces.
Discontinuous responses — illustrative comparison of reported tests
Independent published tests, not conversation history. Gray comparison requirements are not proven missing components. Full historical chatbot messages are unavailable.
1. Same question
Reported test question (excerpt): Are buildings required to accept Section 8 vouchers?
Reported answer A — paraphrase: In one reported test, MyCity said landlords had to accept housing vouchers. The publication does not supply that complete chatbot answer or its qualifications. This panel preserves the reported finding without inventing the missing wording.
Graph callout: One reported answer recognized the obligation. Its actual retrieval path is unknown.
2. Answer
Same reported question, separate tests: Are buildings required to accept Section 8 vouchers?
Reported answer B — paraphrase: In other tests, MyCity said buildings did not have to accept tenants using housing vouchers. The Markup reported that ten staff members asked the same question and all received that negative answer. These are reported outcomes across tests, not an invented later turn in the first conversation.
Graph callout: The same question produced opposing advice, despite published city guidance.
3. Supporting info
GovKM review question — written for this demo: What relationship failed to remain dependable?
GovKM comparison — not a chatbot reply: The reported advice was not consistently aligned with published city guidance. A blanket denial omitted the protection that applies to covered rentals. The conflicting answers are documented, but the public evidence does not establish MyCity’s internal technical cause.
Graph callout: Published guidance and delivered advice were not consistently aligned.
Final comparison graph; shared positions and relationships, with different continuity status:
GovKM continuity analysis and proposed solution
MyCity’s reported answers did not reliably reflect the city’s governing guidance. GovKM describes this as a continuity problem: relevant information existed, but its meaning and authority did not remain dependable in the advice delivered.
OrgAI’s published purpose is to preserve organizational context, evidence, and authority over time. For this example, the intended improvement is a qualified answer that reflects the applicable city guidance and makes its limits understandable.
The graphs show the same public-information task. The highlighted OrgAI contribution represents the intended benefit of continuity. The historical failure is not attributed to OrgAI’s absence, and the public evidence does not establish MyCity’s technical root cause.
This is a simulated solution, not a verified native GovKM result or evidence that OrgAI has corrected the deployed MyCity system. The example describes the desired response, not the internal methods used to produce it.
Continuous response — simulated OrgAI contribution
Illustrative exchanges describing intended outcomes. This is not a live OrgAI service or a technical implementation demonstration.
1. Same question
Prompt: What information matters when answering this NYC housing-voucher question?
Response: The answer needs the relevant NYC housing guidance and the circumstances of the rental. Coverage and exceptions matter. OrgAI’s intended role is to support continuity of that organizational context so the question is considered with the information that gives it meaning.
Graph callout: OrgAI’s intended contribution: relevant guidance and context remain connected.
OrgAI contribution: OrgAI is designed to preserve the context that makes organizational information useful. Here, the question concerns NYC housing guidance and the circumstances in which it applies.
2. Answer
Prompt: What would a fuller, qualified response explain?
Response: For a rental covered by NYC’s source-of-income protections, a landlord cannot reject an applicant solely because they use a Section 8 voucher. Exceptions apply to certain properties, so coverage must be established. Other lawful screening considerations remain separate; this does not mean every applicant must be approved.
Graph callout: The intended result is a qualified response consistent with the applicable guidance.
OrgAI contribution: OrgAI’s published proposition is better continuity of evidence, authority, and context for people and AI. The highlighted graph illustrates the intended improvement in the information available for this response.
3. Supporting info
Prompt: What should remain useful when the question is asked again?
Response: The relevant guidance, circumstances, qualifications, and any later changes should remain understandable. If the rules or circumstances differ, the answer may also differ. The intended continuity outcome is informed future use, with uncertainty explained and human judgment retained.
Graph callout: OrgAI’s intended contribution: organizational knowledge remains useful over time.
OrgAI contribution: OrgAI is designed to support usable organizational knowledge across time. GAIB makes those relationships visible. This example illustrates that public-facing purpose without describing internal methods.
Final comparison graph; shared positions and relationships, with different continuity status:
Highlighted relationships identify the intended OrgAI contribution; they do not disclose internal methods or demonstrate an operating service.
Sources
- NYC Comptroller, MyCity audit, December 30, 2025
- Reported housing-voucher tests — The Markup, March 29, 2024 — Observed responses across separate tests; full transcripts are not supplied.
- Governing guidance — NYC Commission on Human Rights — Lawful source of income, protected properties, and exceptions; reviewed October 28, 2026.
- Official closure notice — NYC.gov — Confirms that the chatbot beta test ended.
- Shutdown circumstances — The Markup, January 30 / February 4, 2026 — Reports both usability concerns and the budget context.
OrgAI and organizational continuity
OrgAI is GovKM’s specified architecture for organizational continuity. It is designed to keep evidence, authority, context, decisions, and records connected over time while organizations continue using their existing systems and AI.
- Question: Can a landlord reject an applicant because they use a housing voucher?
- Circumstances: New York City, whether the rental is covered, and applicable exceptions.
- Public guidance: NYC Commission on Human Rights information on lawful source of income.
- Qualified answer: For a covered rental, voucher use alone does not justify rejection.
- Limits: Coverage and other lawful screening considerations still matter.
1. Information stays in context
A useful answer needs the circumstances that give the information meaning. OrgAI is designed to preserve organizational context across people, systems, and time.
2. Evidence and authority remain visible
People should be able to understand where information comes from and which authority supports its use. OrgAI’s public proposition is continuity that makes organizational knowledge more explainable and trustworthy.
3. Human judgment remains essential
OrgAI does not determine organizational truth. People remain responsible for judgment and authorization. The intended outcome is better-informed use of organizational knowledge, not an automatic guarantee that every AI answer is correct.
4. Knowledge remains useful over time
Organizational knowledge should remain understandable beyond the original conversation, project, or personnel change. OrgAI is designed to support that continuity, and GAIB makes the relationships visible for people to explore.
In this example, the intended improvement is a response that reflects the relevant guidance and circumstances, explains its qualifications, and remains useful for future questions.
Illustrative scenario. The diagrams show intended outcomes, not a live OrgAI service or a demonstration of its internal implementation.
Continue Exploring
When Government Systems Are Replaced, What Happens to Their Authority?
The Practice Form Failed. The Agent Used the Real One.
