Iveda a global leader in AI-driven video analytics and smart city technologies, today announced a major advancement within its IvedaAI platform: real-time zero-shot AI detection powered by natural language prompts. For the first time, retailers and security operators can type a single word or phrase — "shoplifting," "suspicious behavior," or "graffiti" — and instantly activate a custom AI detection model that begins analyzing live video feeds on the spot, with no model training, data labeling, or deployment lag required.
Real-time zero-shot detection leverages advanced Vision Language Models (VLMs) alongside Iveda's decade of pre-trained object detection models.
Being evaluated with one of world's largest fast-fashion retailers (thousands of stores across 90+ countries).
Two deployment options: Cloud-Connected Mode or Local Secure Deployment (Cosmos-Reason engine).
Next-generation Cosmos-Reason-2 engine shipping within coming quarter.
Detects shoplifting, suspicious behavior, loitering, graffiti, smoking in restricted zones, fighting, and more.
Available as add-on annual license within IvedaAI platform.
"This is the greatest leap forward yet in real-time AI video analytics," said David Ly, CEO and Founder of Iveda. "For years, building an AI detection model meant collecting data, labeling it, and waiting. Now, a retailer can type 'shoplifting' and, in seconds, have a fully functioning AI model running live across their camera network. That's never been possible before — and the accuracy of what we're detecting is remarkable."
The announcement marks a fundamental shift in how AI-powered video surveillance works. Traditionally, building a detection model required extensive datasets, weeks of training, and significant technical resources. With IvedaAI's new capability, that process is compressed to seconds. The moment a user submits a prompt, a custom AI model is constructed in real time and immediately applied to live or recorded video — turning a natural language instruction into an active, accurate detection engine.
Shoplifting has long been one of the most difficult behaviors to address through traditional AI surveillance. It manifests in countless forms — concealment in bags, pockets, clothing, or carts — across a wildly diverse range of individuals and store environments. No training dataset could ever completely capture every variation, which made automated detection unreliable at best.
IvedaAI's zero-shot approach sidesteps that problem entirely. By leveraging advanced Vision Language Models (VLMs) alongside Iveda's decade of pre-trained object detection models, the platform can interpret the intent and context behind behaviors, not just the static objects in frame. The capability is already being evaluated with one of the world's largest fast-fashion retailers – with thousands of store locations across more than 90 countries – to address loss prevention at scale.
In early testing, retailers saw immediate results. A prompt of "suspicious behavior" successfully flagged individuals peeking into windows and people repeatedly scanning staff-only entry points — demonstrating the system's ability to detect intent-driven, pre-incident activity that would have been invisible to conventional analytics.
Shoplifting across diverse concealment methods
Suspicious or pre-incident behavior
Loitering in sensitive areas
Graffiti attempts and vandalism indicators
Smoking in restricted zones
Fighting or aggressive conduct
All detections are available immediately upon prompt submission, directly within a retailer's existing IvedaAI dashboard — no new login, no new infrastructure required.
Cloud-Connected Mode: Users can connect to large language model processing for live frame analysis without investing in additional on-site hardware. Ideal for organizations looking for rapid activation.
Local Secure Deployment (Cosmos-Reason Engine): For retailers requiring closed-network environments, Iveda deploys a dedicated on-premise server powered by its proprietary Cosmos-Reason engine. This option ensures zero open ports and fully local processing — delivering the full power of real-time AI analysis without any cloud dependency. Iveda will begin shipping the next-generation Cosmos-Reason-2 engine within the coming quarter, further enhancing inference performance and detection accuracy.
As a platform that has spent more than ten years developing and refining AI object detection—including weapons recognition, license plate identification, and facial recognition—Iveda is uniquely positioned to deploy zero-shot AI at scale. The new capability layers advanced VLM reasoning on top of that proven foundation, enabling detection performance that holds up in low-light, crowded, and complex retail environments.
"We've always been committed to staying at the cutting edge," added Ly. "We're already integrating the latest frameworks from NVIDIA and are continuously future-proofing the platform. What we're announcing today is the result of years of foundational work meeting this new wave of AI capability — and retailers are going to feel the difference immediately."
About Iveda Solutions
Iveda is the provider of global solutions for cloud-based, video AI search and surveillance technologies that protect the people, places, and things that matter the most. Iveda's technology has the power to provide instant intelligence to existing infrastructure, enabling cities and organizations around the world to seamlessly enter the fifth industrial revolution. Iveda operates at the forefront of digital transformation of cities across the world, using IoT platforms with smart sensors and devices developed to aid with use cases surrounding public safety, security, elderly care, energy efficiency, and environment preservation. Headquartered in Mesa, Arizona, with a subsidiary in Taiwan, Iveda is publicly traded on NASDAQ under the ticker symbol "IVDA".
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.