Copy article

AI trained on 68 smuggled specimens can now spot shark fins hidden in airport luggage 95% of the time

ended 09. June 2026

AI trained on 68 smuggled specimens can now spot shark fins hidden in airport luggage 95% of the time, using scanners most travellers walk past without a second thought.

Researchers at Sydney's Macquarie University trained an AI to identify shark fins in airport CT baggage scans with 95% accuracy. Seahorses, 96%. The study, published in Frontiers in Ocean Sustainability, hit 92% overall across three trafficked marine species, including sea cucumbers at 86%. The algorithm learned from specimens sourced from actual seizures, hidden beneath clothing, wrapped in foil and stuffed inside children's toys.

Older X-ray systems give security staff a flat image. CT scanners produce a 3D image they can rotate and slice through, so a shark fin tucked behind a laptop battery actually looks like a shark fin.

No new hardware required. The false positive rate sits at 13%, so human officers and sniffer dogs remain essential. This is not a replacement. It is a detection layer that didn't exist before.

  • Roughly 100 million sharks are killed every year, driven by the illegal fin trade (UN). Removing apex predators collapses local fish populations and destabilises marine food chains.
  • The economic damage is estimated at $1 trillion to $2 trillion per year (World Bank).
  • The UNODC and WWF note that wildlife smuggling profits fund armed militias, insurgencies and regional instability.

We'd like your views:

  • If this AI gets deployed on existing airport scanners, could it shift the economics of marine wildlife smuggling, or will traffickers simply find routes that bypass airports?
  • Ivory and rhino horn get the headlines. Marine wildlife trafficking is worth billions but barely registers publicly. Does a working detection tool change that, and does it matter if the public never notices?
  • This is AI doing something concretely useful. No jobs lost, no surveillance overreach, no hype. If stories like this can't cut through the doom narratives, does growing public resistance to AI end up blocking the beneficial uses too?
  • The algorithm hit 95% accuracy on shark fins from just 68 specimens. What happens when the dataset scales to thousands? Could airport smuggling of marine wildlife become functionally unviable?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
A 92% accurate detector built on existing airport hardware is exactly the kind of AI use case that should be front and centre, and it won't be.

The AI industry spent the last three years selling automation fantasies, surveilling workers, slashing jobs and generating slop, then acted surprised by the massive public backlash. Every over-hyped product launch, every quietly deployed facial recognition system, every chatbot hallucination passed off as expertise has made it harder for something like this to land. The researchers at Macquarie University have done clean, grounded work. No jobs threatened, no privacy violated, no breathless claims about changing everything. Just a tool that spots a dried seahorse in a suitcase better than a human eye can. If that story can't compete with the latest doom cycle, the industry built that barrier itself.

The real question is how a university press office outmarkets hyperscaler companies spending billions to own every AI headline on the planet.
Copy

Editor
Samuel Mather-HolgateVerified
Managing Director & IFA at Mather and Murray Financial
From an investment perspective, this is exactly the kind of AI story that matters. A real-world productivity tool layered onto existing infrastructure. If software can turn airport CT scanners into a far better filter for wildlife trafficking, it shows why AI remains such a powerful long term investment strategy. The opportunity is not just in chatbots and LLM's, but in specialist applications across security, healthcare, logistics and compliance. As datasets scale from dozens to thousands, accuracy should improve further, making these systems more commercially valuable and harder to ignore. That is why AI still looks highly investable to me: it is moving from novelty to utility. The winners may not just be the headline AI giants, but the firms applying it to solve expensive, specific problems at scale.
Copy

Editor
Katrina YoungVerified
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
The headline is 95%. The story is whether 95% survives contact with a real airport. Underneath the number sit 68 physical specimens. To expand the dataset, the team used synthetic threat-image projection techniques, inserting digital examples into baggage scans to simulate concealment scenarios. That is the part worth scrutiny. Detection models built in controlled conditions can behave differently in live environments: cluttered luggage, novel concealment methods and species the system has never seen. The authors acknowledge the limited sample size may affect how well the model generalises to real-world deployments. The 13% false alarm rate means officers still clear every alert by hand. None of this makes the work less valuable. It makes the evidence question the real one. Before governments scale systems like this across borders, the challenge is proving they work in the wild. A percentage is a claim. A control is a result that survives reality.