Industry Alerts

A retired US federal judge recently made an unsettling admission. Reviewing a legal brief, he found the cited authorities so persuasive that he looked them up to read more — and discovered they didn’t exist. They had been invented, whole cloth, by an AI tool. “Directly put,” he wrote, the AI “affirmatively misled me,” and he had “almost” written the fabricated cases into a binding court order.

That story isn’t about lawyers. It’s about a deeper truth that everyone in cargo, freight and logistics needs to sit with as AI sweeps through the industry.AI creates its greatest risks precisely when it sounds most convincing. It doesn’t fail with an error message. It fails with total confidence — a clean, fluent, authoritative-sounding answer that happens to be wrong. And in a business built on documents, declarations and data, a confidently wrong answer can be very expensive.

THE BOOM IS REAL:

Let’s start with the good news, because it’s genuine. AI has moved from pilot projects to everyday operations across shipping and logistics, and the gains are measurable, not hypothetical:

  • Route and voyage optimisation. AI models weigh weather, currents, port congestion and fuel burn to plot more efficient voyages — cutting cost and emissions.
  • Predictive maintenance. McKinsey estimates predictive maintenance can cut maintenance costs by up to 40% and downtime by up to 50% — increasingly treated as a baseline, not an experiment.
  • Document automation. Generative AI turns the industry’s sea of unstructured emails, PDFs and phone calls into structured, actionable data — quoting, bookings and back-office tasks that used to eat hours.
  • Visibility and ETAs. AI-driven tracking platforms give dynamic, real-time arrival estimates and flag disruptions before they bite.
  • Cargo safety screening. As we’ve covered before, the World Shipping Council’s new AI screening tool scans millions of bookings to catch misdeclared dangerous goods before they’re loaded.

The headline economics are striking Microsoft research cited across the sector suggests AI-enabled innovation could cut logistics costs by around 15%, improve inventory levels by roughly 35%, and lift service levels significantly. Little wonder adoption is racing ahead — customs operators using AI alone were projected to jump from 23% in 2024 to 45% in 2025.

 

Where AI genuinely helps the cargo-claims world
  • Faster, more consistent processing of claims paperwork — bills of lading, surveys, correspondence — pulled together in minutes, not days.
  • Pattern-spotting across large claims datasets to flag fraud indicators, recurring loss causes, or high-risk lanes.
  • Drafting support — first-pass summaries, timelines and letters that a human expert then refines.
  • Triage — sorting a busy inbox of notifications by urgency and complexity so nothing slips.
Used well, AI doesn’t replace the claims expert. It clears the busywork so the expert can focus on judgement.

 

Here is the part the glossy adoption headlines tend to skip. The same generative AI that drafts a flawless-looking summary can also “hallucinate” — produce information that is fabricated, misattributed or subtly distorted, while sounding entirely authoritative. The scale of this is no longer theoretical.

 

The hallucination problem, in numbers
  • As of early 2026, an estimated 712 court decisions have now addressed AI-hallucinated content — and roughly 90% of them were written in 2025 alone. (AI Hallucination Cases database, HEC Paris)
  • In one US case, 12 of 19 legal citations in a single filing were fabricated, misleading or unsupported.
  • Sanctions have escalated from a few thousand dollars to fines of $10,000–$31,000+ — plus bar referrals, struck filings and public reprimands.
  • A judge admitted he was “almost” misled into adopting fabricated AI citations in an official order.

 

Translate that out of the courtroom and into a freight office, and the risk is identical in shape. An AI tool asked to classify cargo for a customs declaration, summarise a liability position, or interpret a clause in a convention can produce an answer that is fluent, specific, and wrong — and a busy operator under time pressure may never think to check it.

“AI doesn’t fail with an error message. It fails with total confidence.”

The logistics-specific failure modes are already being documented. Customs-technology specialists warn that AI tools struggle with the grey areas and constant changes in customs rules, which vary by jurisdiction and update frequently — and that the consequences of a wrong classification are real: shipments held at borders, or goods misdeclared, with all the cost and liability that follows. Industry analysts describe the “garbage in” problem bluntly: feed an AI distorted or biased data and it learns the wrong lessons, then makes confident decisions that can throw a supply chain off balance.

The US National Security Agency went further in March 2026, warning that AI itself now behaves like a supply chain where bad data, “data poisoning” or a compromised model at any layer can quietly corrupt forecasts, ETAs, routing and compliance reporting downstream.

WHY THIS MATTERS FOR CARGO CLAIMS AND LIABILITY

For a claims and liability business, the unchecked-AI problem isn’t an abstract IT concern — it’s a live source of disputes:

  • Misclassification and misdeclaration. An AI that assigns the wrong tariff code or mishandles a dangerous-goods entry can put a shipper on the wrong side of a compliance failure — or, as we’ve written, contribute to a catastrophe.
  • Reliance and responsibility. Courts are signalling clearly that using AI does not transfer the blame. The duty to verify stays with the human professional. “The AI told me so” is not a defence.
  • Audit trails and explainability. When a decision is challenged, can you show how it was reached? “Black box” AI that can’t be explained becomes a liability in any investigation or audit.
  • Data confidentiality. Feeding sensitive shipment, pricing or claims data into the wrong AI tool can expose commercially confidential or personal information.

GETTING IT RIGHT: HOW TO USE AI WITHOUT GETTING BURNED

The lesson from the courtroom isn’t “don’t use AI.” It’s “never trust it blindly.” The organisations getting real value are the ones who treat AI as a fast, tireless junior assistant — enormously useful, but whose work is always checked before it goes out the door. A practical playbook:

  • Verify every source. If AI cites a case, a regulation, a figure or a clause, confirm it against the primary source before relying on it. Treat unverified AI output as a draft, never a fact.
  • Keep a human in the loop. Use AI to draft, triage and summarise — but let an experienced person make the judgement call, especially on classification, liability and compliance.
  • Choose the right tool. General-purpose chatbots hallucinate more than purpose-built tools with verified data sources. Match the tool to the stakes of the task.
  • Mind the data going in. Don’t feed confidential cargo, pricing or personal data into open AI tools. Apply access controls and know where your data goes.
  • Keep audit trails. Log what the AI was asked, what it produced, and who checked it. If a decision is ever challenged, that record is your defence.
  • Train your people. The single biggest protection is a team that understands AI can be confidently wrong — and knows it is their job to catch it.

COLLYERS VIEW

AI is a genuine step forward for cargo handling and freight operations. Used well, it makes the whole chain faster, cheaper and safer. But the same quality that makes it powerful — its fluent, confident output — is exactly what makes blind reliance dangerous. In a business where a single wrong figure on a declaration or a single fabricated authority in a position can decide a claim, verification is not optional it’s the whole job.

That is, in a sense, what Collyers has always done: check what’s really in the box, what the documents really say, and where the chain really failed. AI changes the tools. It doesn’t change the discipline if anything, it makes that discipline more valuable than ever.

Navigating cargo claims in an AI-driven supply chain?

Collyers provide specialist human expertise to handle marine cargo, air freight and road freight liability claims accurately, and from first notification to final recovery.
Talk to our claims team about how we can help visit www.collyers.com or get in touch directly.

 

Sources & Further Reading
  • McKinsey & Microsoft research, cited in Reform / Closeloop logistics reviews — predictive maintenance and AI cost/efficiency gains.
  • Freightos FreighTech 2025; ARC Advisory Group — AI freight-tech adoption and supply-chain risk.
  • AI Hallucination Cases database, Damien Charlotin (HEC Paris); Bloomberg Law; Ars Technica — ~712 decisions, escalating sanctions, the “almost” order.
  • Mouse & Bear / ITBrief — AI customs-classification errors (23%→45% operator adoption 2024–25).
  • US National Security Agency — AI supply-chain risk guidance, March 2026.
This article is provided for general information and does not constitute legal, compliance or claims advice.

 

Collyers Industry Alerts – providing practical insights into developments affecting global cargo claims and freight liability risk.

Industry Alerts

Three Letters

Who bears a cargo loss under FOB, FCA, CIF or DAP? Understand Incoterms® 2020 risk transfer, insurance and cargo claims, plus the Incoterms® 2030 revision now under way.

Read more
Industry Alerts

Three Degrees From Disaster

Reefer and cold chain cargo failures account for around 30% of container-vessel cargo claims. Why pharma and food losses are uniquely complex, the five ways cold chains break, and how to build a claim that survives. From Collyers.

Read more