Skip to content
NB Tech NewsSubscribe

News1 min read

UNB Researcher's 'Trust Passport' Would Stamp Every AI Reply with Its Risk Profile

A Canadian Institute for Cybersecurity poster proposes a per-response risk snapshot for enterprise LLM pipelines, flagging unreliable retrieval, sensitive-data exposure, risky tool use, and weak factual support.

Retrieved documents feeding an LLM reply bubble, four coloured risk checks below it, and a trust passport card stamped onto the reply

What if every AI-generated reply came with its own Trust Passport — a stamp-sized summary of how much confidence you should actually place in it?

That's the idea behind a poster from UNB's Canadian Institute for Cybersecurity researcher Vidushi Agarwal, set to appear at the PST Cybersecurity Summit 2026 in Fredericton on Aug. 20.

The problem: confident, unverifiable answers

Enterprises are wiring large language models into pipelines that retrieve documents, call tools, and answer directly to employees and customers. LLMs deliver those answers with uniform confidence — whether the underlying retrieval was solid, the tool call was risky, or the model is improvising. For a business deciding whether to act on an AI answer, that flat confidence is the problem.

The approach: a risk snapshot per response

The Trust Passport concept gives each AI response a clear snapshot of potential risks across four signals:

  • Unreliable retrieved information — did the sources the model leaned on actually support the answer?
  • Sensitive-data exposure — did the pipeline surface confidential or personal data it shouldn't have?
  • Risky tool use — did the model invoke tools or actions that carry consequence?
  • Weak factual support — is the answer grounded, or is it plausible-sounding filler?

By bringing those signals together, the passport lets users quickly gauge how much confidence a response deserves — and, the poster argues, make better-informed decisions when working with LLMs.

Why it matters

"Trust but verify" is the operating principle of most security-conscious AI deployments, and verification is currently manual and ad hoc. A standardised per-response risk label is the kind of infrastructure that could make enterprise LLM adoption safer — and it's a research direction squarely in the wheelhouse of the institute that anchors Fredericton's security cluster.

Sources

ChronologyCIC· 11 stories since Aug 4, 2026

  1. Mastercard's Canada Security VP Keynoted CIC's Fredericton Summit — Then Toured the IoT Lab It Funds
  2. UNB Researchers Threat-Model the Protocols That Let AI Agents Act
Chronology · show all 11 stories
  1. UNB Researchers Build a Practical Roadmap for the Post-Quantum Migration
  2. UNB Poster Weighs the Privacy Cost of Federated-Learning Aggregation
  3. UNB Researchers Deploy Collaborative AI Agents to Guard Power-Substation Networks
  4. UNB Researchers Build a Mixture-of-Experts Defence Against Adversarial AI Attacks
  5. UNB Researchers Make AI Malware Detection Explainable with a 'Dual Prototype' Framework
  6. UNB Researcher Takes a Geometric Axe to Optimization's Symmetry Problem
  7. UNB Researcher's 'Trust Passport' Would Stamp Every AI Reply with Its Risk Profile (this story)
  8. National Cyber Agencies and Big Banks Head to Fredericton for UNB's Cybersecurity Summit
  9. Cybersecurity Pioneer Ali Ghorbani Named to the Order of New Brunswick

Related coverage

Tags