DSME Construction v. U.S. Army (ASBCA): Korean Contractor’s Surreply Struck for 16 Non-Existent Cases; Dismissal Denied

Case at a Glance
Court: Armed Services Board of Contract Appeals (ASBCA)  | 
Citation: ASBCA No. 63878, March 13, 2026  | 
Outcome: Surreply struck for 8 non-existing cases (out of 27 problematic citations); dismissal of $100M+ contract appeal denied; motion to amend surreply denied; “lack of candor” finding  | 
Issue: AI hallucinations in Korean contractor’s ASBCA surreply — 10 non-existent cases across 3 briefs; dismissal denied as too severe
ElementDetail
AppellantDSME Construction Co., Ltd. (Korean contractor, represented by Korean counsel)
RespondentU.S. Army (preventative maintenance services at Camps Humphreys and Yongin, Republic of Korea)
AI Conduct27 problematic citations in surreply (16 non-existent); 10 non-existent cases total across 3 briefs; “lack of candor” about how errors arose
OutcomeSurreply struck; dismissal denied; amended surreply motion denied; “lack of candor” finding — appeal continues on remaining briefs
DateMarch 13, 2026

Background: A Default Termination, ASBCA Appeal, and a Problematic Surreply

DSME Construction Co., Ltd., a Korean contractor, challenged the U.S. Army’s default termination of a contract for preventative maintenance services at military facilities in South Korea. The parties proceeded under ASBCA Board Rule 11 — a decision-on-the-briefs format without live hearings — filing opening, responsive, and surreply briefs in a structured schedule. After the government reviewed DSME’s December 1, 2025 surreply, it identified 27 problematic citations including 16 to cases that do not appear to exist. The government characterised these as “the hallmarks of generative artificial intelligence (AI) hallucinations.”

DSME’s counsel — a Korean IP and law firm — did not address whether AI was used, but apologised and sought to file an amended surreply, attributing the errors to “a fundamental breakdown in the appellant’s citation-verification and proofreading process.” When the government pushed for dismissal of the entire appeal, DSME admitted that 8 non-existing cases appeared in the surreply and asked for the lesser sanction of striking that brief. The government’s review also found 5 additional problematic citations in DSME’s opening brief.

The AI Issue: 10 Non-Existent Cases Across Three Briefs + Lack of Candour

The ASBCA found that DSME had cited 10 non-existing cases across three separate briefs (opening, surreply, and at least one other). The Board found DSME’s explanation for how those non-existing cases found their way into the briefing to be lacking in candour — a finding that carries significant weight before a quasi-judicial tribunal. The Board denied the motion to file an amended surreply specifically because of this “lack of candour” combined with the volume of citation errors and the deterrence rationale.

The government’s request for dismissal of the entire appeal with prejudice was denied. The Board found the government failed to show this was “contumacious or contemptuous conduct” sufficient to justify the “drastic sanction of dismissal” — described as “the severest sanction the Board can issue.” This is a significant calibration point: 10 non-existent cases across three briefs does not reach the threshold for terminating sanctions at the ASBCA.

Holdings

  1. Surreply struck. DSME’s December 1, 2025 surreply was struck for containing 8 admitted non-existing case citations.
  2. Dismissal denied. The government’s motion to dismiss the entire appeal with prejudice was denied — the Board found the standard for “contumacious or contemptuous conduct” necessary for this sanction was not met.
  3. Amended surreply motion denied. Despite DSME’s offer to file a corrected surreply, the motion was denied in view of the “lack of candour” regarding how the errors arose and to deter future AI-related citation failures.
  4. Appeal continues without surreply. The appeal will proceed on the remaining briefs — opening brief and responsive briefs — with the corrective value of the surreply lost.

“In view of appellant’s admission that three of its briefs in this appeal together cite ten non-existing cases, combined with what we view as appellant’s lack of candor regarding how those non-existing cases found their way into its briefing, and to deter repetition of future such conduct — AI-related or not — we deny appellant’s motion for leave to file an amended surreply.”

— Administrative Judge Timothy P. McIlmail, ASBCA No. 63878, March 13, 2026

India Angle: AI Hallucinations in International and Cross-Border Government Contract Disputes

The DSME case involves a Korean contractor with Korean counsel challenging a U.S. Army contract — a genuinely international AI hallucination case. India faces a similar dynamic: Indian contractors working on defence or infrastructure projects for foreign governments, or international contractors working on Indian public procurement projects, often use AI tools in cross-jurisdictional brief drafting. The verification duty is heightened in cross-border contexts because counsel may be less familiar with the foreign jurisdiction’s case law.

Relevant Indian Law

  • Defence Procurement Procedure 2020 / DAP 2020: India’s defence procurement involves both Indian and foreign contractors. Contract disputes go to arbitration under the India-specific dispute resolution clauses. AI-assisted briefs in ASBCA-equivalent Indian proceedings (DRDO, MoD arbitrations) require the same citation accuracy standards.
  • Arbitration and Conciliation Act 1996, Section 31A: Arbitral tribunals can impose costs for unreasonable conduct in proceedings. AI-generated fabricated citations in arbitral submissions could ground a cost order against the submitting party.
  • Legal consequences for foreign counsel: Korean counsel Yong Eui Song appeared before the ASBCA. In India, foreign lawyers cannot appear without Indian co-counsel. Any Indian advocate who co-signs a submission containing AI-generated fabricated citations is personally responsible under the BCI Rules, regardless of who actually drafted the document.

Three Practical Tips

  1. In cross-border cases, never use AI to generate citations from an unfamiliar jurisdiction’s case law without independent verification. DSME’s Korean counsel generated citations to U.S. federal contract law cases that did not exist. AI tools are particularly unreliable in cross-jurisdictional citation contexts because the training data for foreign law is sparse. Always use an authoritative database for the target jurisdiction.
  2. Lack of candour about how errors arose is independently damaging. DSME’s failure to clearly explain whether AI was used, and how the errors arose, led to the “lack of candour” finding that directly contributed to denial of the amended surreply motion. When AI errors occur, full and transparent disclosure of how they arose is the safest path.
  3. Dismissal as a sanction has a high threshold — but losing your surreply brief is a significant tactical loss. The ASBCA declined to dismiss DSME’s entire appeal, but striking the surreply brief leaves the appellant without the last word in a brief-based proceeding. Indian arbitration practitioners should understand that “lesser sanctions” like brief-striking can still be case-determinative.

Quick Takeaways

  • 27 problematic citations in a single brief (16 non-existent) is the highest count documented in ASBCA/government contract AI hallucination case law.
  • The ASBCA has now decided that 10 non-existent cases across three briefs does not reach the threshold for dismissal — providing a calibration point for what “contumacious” conduct means in government contract appeals.
  • Lack of candour about AI use — failing to clearly disclose whether AI was used and how citation errors arose — is an independently sanctionable factor separate from the citation errors themselves.
  • An amended surreply offer does not guarantee a second chance: if the tribunal finds the explanation for original errors insufficient, it can deny amendment as a deterrence measure.
  • In India: cross-jurisdictional AI citation errors in defence or infrastructure contract arbitrations carry both case-level risks (brief-striking) and personal-level risks (BCI liability for the Indian co-counsel who signed).

What Our Clients Say

Chat on WhatsApp Call Now
Scroll to Top