From the Local Government edition of October 23, 2026
Denver auditor's AI governance audit: Technology Services agrees to nine fixes, from an AI inventory to defining AI security incidents, with most due in late 2027
City and County of Denver, Office of the Auditor and Technology Services · City IT governance and audit · Denver, Colorado, United States
- Publisher
- Denver Auditor's Office
- Original publication
- Agency response dated 6 October 2026; audit released 23 October 2026 (Denverite)
- Source retrieved
- 2026-10-24
- Event date
- 2026-10-23
What happened
Denver's auditor found that the city's Technology Services AI-governance framework needs improving, and the department agreed to all nine recommendations. They include a formal AI risk-management process, defining trustworthy AI using the NIST framework, tracking which vendors embed AI in their products, a complete AI inventory with an approved and denied list visible to all staff, organisation-wide AI data-protection controls, post-deployment performance reviews, and a written definition of AI security incidents such as prompt injection. Most target dates fall between October and December 2027; data-protection controls are due by 31 December 2026. Press coverage on 23 October said the airport has no AI framework at all.
Why it matters
Why now: the audit was released on 23 October. A large city that already runs resident-facing AI was told by its own auditor that basic governance plumbing, such as an inventory, vendor AI mapping and incident definitions, is not yet formalised.
Evidence and measured results
Official agency response listing each recommendation, agreement and target date. The audit's underlying test results were not in the opened document.
Limitations and uncertainty
Response document only; findings detail and airport response not opened. Long target dates mean gaps may persist through 2027.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-23; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- The city runs AI but lacks a formal risk process, complete inventory, vendor AI mapping and an AI incident definition, and its auditor is now tracking fixes.
- Stakeholders
- Chief AI and information officer
- City auditor
- CISO and data protection team
- Enterprise architecture and AI director
- Procurement
- Airport business technologies leaders
- Discovery questions
- Is your AI inventory in the asset system of record, and does it include vendor-embedded AI?
- Which AI systems make operational decisions, and what metrics do stakeholders see?
- Have you defined an AI security incident?
- When is your next IT audit?
- Value hypothesis
- Bounded: audit-ready governance tooling and processes could shorten remediation, but this source gives no cost or effort estimates.
- Potential engagement
- An audit-readiness review mapping current AI controls against the nine Denver recommendations, with a prioritised remediation plan.
- Unsupported claims
- That any Denver AI tool delivered or failed to deliver benefits
- Remediation cost or timeline estimates
- That NIST alignment satisfies an auditor
Pre-sales engineering
Role takeaway
- Fit
- Cities with several AI tools in production and a central IT or AI office.
- Architecture and integration
- Inventory in the IT asset system of record; vendor intake captures AI features and system-to-system interactions; AI DLP across the tenant; post-deployment monitoring by an AI working group.
- Prerequisites
- Asset management system
- Vendor intake workflow
- DLP platform
- Incident management process
- Constraints
- Colorado and city regulations still evolving
- Target dates of late 2027
- DLP synchronisation issues noted
- Security
- Define AI incident classes (prompt injection, model extraction, cyberattacks, data exfiltration, unauthorised access) and verify data-protection controls periodically.
- Proof of value
- Reconcile the AI inventory against procurement records
- Test AI DLP rules on sample sensitive data
- Run a tabletop exercise on a prompt-injection incident
Delivery
Role takeaway
- Work
- Implement nine agreed audit recommendations covering risk management, regulation monitoring, trustworthy AI definition, output-impact policy, risk mapping, inventory, data protection, performance review and incident definition.
- Dependencies
- Auditor follow-up schedule
- Technology Services staffing
- Mayor's Office legislative team
- Vendor cooperation on AI disclosure
- Ownership
- Denver Technology Services, with named points of contact per recommendation.
- Skills and adoption
- Staff need clarity on allowable AI use; an approved and denied AI list visible to all employees supports that.
- Governance checkpoints
- AI data-protection controls by 31 Dec 2026
- AI incident definition by 30 Apr 2027
- Risk management, inventory and output policy by late 2027
- Auditor follow-up review
- Acceptance criteria
- Proposed: 100% of AI systems and AI-embedding vendors listed in the asset system
- Proposed: approved and denied AI list published to all staff and updated on a set cadence
- Proposed: post-deployment review completed for each production AI system
- Risks
- Target dates slip beyond 2027
- Inventory misses vendor-embedded AI
- Airport remains without a framework
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Asks for AI-risk mapping in vendor intake, including how AI systems interact with each other, and for the AI inventory to sit in the asset system of record.
Governance
Who approves, reviews and stays accountable for outcomes?
Treats NIST AI RMF 1.0 as the minimum definition of trustworthy AI and requires performance metrics for AI that makes operational decisions to be available for stakeholder oversight.
Security and privacy
What data, permissions and controls need testing?
AI data-loss-prevention controls are being deployed but have synchronisation issues; AI security incidents (prompt injection, model extraction, data exfiltration) are not yet defined in policy.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Staff need clearer rules on allowable AI use; the audit asks for an approved and denied AI list visible to every employee.
Procurement
What should contracts, pricing and exit terms secure?
Vendor intake must capture AI features embedded in products and how AI systems interact, so procurement becomes an AI risk control point.
Operating model
Which teams own the service once it runs?
Assigns named Technology Services owners to each recommendation and gives an AI working group responsibility for post-deployment review.
Publication history
- 2026-10-23Local Government · Issue 487 resources
Stable resource ID: denver-co-auditor-ai-governance-audit-nine-recommendations-20261023