AI oversight with the audit committee, but no review procedure and no test data
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Oversight of AI risks is assigned to the audit committee, but there are only principles on the AI policy and no described approval procedure.
What is documented
- The 2026 proxy statement gives the audit committee primary responsibility for risks relating to cybersecurity, AI and data privacy1 2
- A body from technology, legal, HR, sales, safety and procurement steers the phased use of AI under an internal policy1
- The code of conduct allows AI tools only in line with the policy, prohibits automated decisions and requires human review3
- The Values Line received 1,239 reports in 2025; the audit committee reviews them quarterly2
The case against
Governance not disclosed
Accountability
- No AI training, metrics or external audit of AI governance documented
- No grievance channel for people affected by AI systems such as employees or truck drivers
Sources for Responsibility (4)
- Union Pacific Corporation / U.S. SEC: Proxy Statement 2026 (DEF 14A) for the annual meeting of 2026-05-14 · 25 March 2026 · primary source
- Union Pacific Corporation: 2025 Building America Report · 31 December 2025 · primary source
- Union Pacific Corporation: The How Matters – Code of Conduct · 10 September 2026 · secondary source
- Union Pacific Corporation: UP Chat (post by Rick Holmes, Chief Security Officer-Technology) · 5 February 2024 · secondary source
2 · Safety
Item 1C documents a concrete cyber programme; the reliability of AI in rail operations is only self-described.
What is documented
- Item 1C describes audit committee oversight, a CISO organisation, third-party reviews, penetration tests and a 2025 tabletop exercise8
- TSA cybersecurity requirements are implemented and validated; no material incidents are reported8 10
- Union Pacific states a SOC 3 certification, attack simulations and security keys9
- The internal AI assistant runs as a secured version in the company network, with a ban on confidential inputs and automated decisions11
- According to Union Pacific, track inspectors decide on machine vision findings1
The case against
Reliability not evidenced
Court cases
Sources for Safety (15)
- Union Pacific Corporation: AI-powered vision inspects track · 22 May 2026 · secondary source
- Union Pacific Corporation: Point protection technology · 25 November 2025 · secondary source
- Union Pacific Corporation: What data shows about modern track inspections and rail safety · 8 September 2026 · secondary source
- Federal Railroad Administration (U.S. DOT): Decision Letter Docket FRA-2025-0059 (waiver request of the Association of American Railroads) · 5 December 2025 · primary source
- FreightWaves: Federal suit alleges AI system injured Union Pacific employee · 25 April 2024 · secondary source
- Cowboy State Daily: Wyoming Railroad Engineer Sues Union Pacific Over AI Train Driver · 22 April 2024 · secondary source
- Cowboy State Daily: Wyoming Man Who Says AI Driver Wrecked Train Settles With Union Pacific · 10 April 2025 · secondary source
- Union Pacific Corporation / U.S. SEC: Form 10-K for the fiscal year ended 2025-12-31 · 6 February 2026 · primary source
- Union Pacific Corporation: 2025 Building America Report · 31 December 2025 · primary source
- Board Cybersecurity: Union Pacific – Cybersecurity Incident Tracker · 19 May 2026 · secondary source
- Union Pacific Corporation: UP Chat (post by Rick Holmes, Chief Security Officer-Technology) · 5 February 2024 · secondary source
- IDTechWire: Judge Certifies 22,000-Driver Biometric Class Against Union Pacific · 24 August 2026 · secondary source
- Leagle: Fleury v. Union Pacific R. Co., 729 F.Supp.3d 854 (N.D. Ill. 2024), Case No. 20 C 390 · 10 April 2024 · secondary source
- National Law Review: BIPA Cases: 7th Circuit Rules Change to Illinois Law's Damages Provision Retroactively Limits Defendant Exposure · 10 April 2026 · secondary source
- Progressive Railroading: Union Pacific enhances L.A. cargo safety via tech, infrastructure upgrades · 1 November 2024 · secondary source
3 · Transparency
Union Pacific names areas of use and principles, but no limits, data or error rates of its AI systems.
What is documented
The case against
Sources for Transparency (10)
- Union Pacific Corporation / U.S. SEC: Proxy Statement 2026 (DEF 14A) for the annual meeting of 2026-05-14 · 25 March 2026 · primary source
- Union Pacific Corporation: 2025 Building America Report · 31 December 2025 · primary source
- Trains (Kalmbach Media): Railroads ramp up the use of artificial intelligence · 3 November 2026 · secondary source
- Union Pacific Corporation: AI-powered vision inspects track · 22 May 2026 · secondary source
- Progressive Railroading: Q&A: Union Pacific's Katie Sanders talks technology and the next frontier in predictive railroading · 21 February 2025 · secondary source
- Earnings Whispers: Union Pacific Q3 2026 Earnings Call Transcript · 22 October 2026 · secondary source
- Cowboy State Daily: Wyoming Man Who Says AI Driver Wrecked Train Settles With Union Pacific · 10 April 2025 · secondary source
- Union Pacific Corporation: UP Chat (post by Rick Holmes, Chief Security Officer-Technology) · 5 February 2024 · secondary source
- Union Pacific Corporation: The How Matters – Code of Conduct · 10 September 2026 · secondary source
- ToS Tracker: Union Pacific Privacy Policy (archived version) · 15 September 2019 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: the lawsuit over locomotive control ended in a settlement without a finding, the biometrics case is pending; there is no ruling, regulatory decision or fine.
5 · Controversies (2)
- mediumIn the Fleury class action a class of around 22,400 truck drivers was certified whose fingerprints are alleged to have been collected without written consent; the case is pending. · 24 August 2026 · open
- mediumAn engineer sued, alleging the LEADER control system gave conflicting commands and caused an accident; Union Pacific denies it is AI, and a confidential settlement followed in 2025. · 18 April 2024 · Vergleich
6 · What remains open
- What does the internal AI policy say, and how are AI systems approved?
- What test, validation and error data exist for machine vision, energy management and camera systems?
- What remedy followed the settlement over locomotive control (training, system changes)?
- Which data protection rules apply to AI cameras, drones and locomotive cameras?
- Item 3 of the 2025 10-K and the legal proceedings of the Q3 2026 10-Q could not be read.
7 · How it was assessed
Cut-off date: 3 November 2026 · 42 searches, 33 fetches.
Primary documents read (8)
- Form 10-K FY 2025 (S01, with continued reading from character 100,000; Item 1C read, Item 3 not included)
- Form 10-K FY 2025 XBRL R27 and FilingSummary (S02)
- Proxy Statement 2026 DEF 14A (S03, three retrievals)
- 2025 Building America Report (S04, two retrievals, characters 0–100,000)
- Code of Conduct «The How Matters», as of 2026-09-10 (S05)
- Form 10-Q Q3 2026 (S29, partial retrieval)
- FRA Decision Letter FRA-2025-0059 (S19)
- Senate hearing 2026-06-09, written testimony AAR (S20)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0