Alibaba adapts to the age of AI fakes

China’s e-commerce company is taking on AI counterfeits with a mix of human and advanced AI tech of its own—and getting results.

Counterfeiters have been quick to use AI to make sure their listings avoid detection online, but they may be disappointed in the results on one platform in particular.

In Alibaba, itself an AI pioneer, they are up against a tech heavyweight. The China-based retailer’s open source large language model (LLM) Qwen 2.5 is among the weapons being employed to tackle AI-based infringement head-on.

Alibaba will be taking part in a seminar and roundtable on this topic at the WIPR Trademark and Brand Protection Summit, which takes place on 28 October in San Francisco.

WIPR spoke with the company’s Matthew Bassiur, vice president, head of Global IP Enforcement, to understand how its own LLM was being used to fight the fakes, why image recognition is key, and why ‘responsible AI’ has a place in IP.

WIPR: Infringers are now using AI to generate infringing content—what are you seeing, and how are you adapting your detection methods?

Matthew Bassiur: Alibaba has observed that infringers increasingly use generative AI tools to produce deceptive content—such as synthetic product images, altered brand logos, or misleading descriptions—designed to evade traditional filters or mislead consumers.

These AI-generated assets can mimic authentic listings or create “lookalike” visuals that confuse consumers. To counter this, Alibaba is continuously evolving its detection methods to recognise the subtle signatures of AI-generated content.

Its multimodal deep learning (DL) system integrates textual, visual, and behavioural data to identify inconsistencies that human eyes might miss—such as unnatural image patterns, mismatched metadata, or linguistic anomalies typical of AI text generation.

Large language models (LLMs) like Qwen 2.5 are also being tried for notice and takedown analysis, with over 90% accuracy in some tests. Additionally, Alibaba combines AI detection with real-time traffic and transaction analysis to expose coordinated abuse patterns that might accompany AI-generated listings.

Importantly, human reviewers remain essential to confirm cases and retrain models. By pairing technical innovation with continuous learning and human validation, Alibaba continues to monitor evolving infringement tactics and ensures its monitoring systems can adapt rapidly to emerging AI-enabled threats.

Can you tell me more about the technology you developed to detect AI-generated misleading images? How does it work?

Alibaba has developed advanced image recognition and semantic analysis technologies to detect AI-generated or manipulated images that mislead consumers or infringe trademarks.

These systems analyse both visual and textual elements in product listings to identify subtle signals of synthetic or altered imagery.

Optical Character Recognition (OCR) scans hundreds of millions of images daily, detecting embedded text, brand names, and logos. Semantic recognition algorithms then assess whether the language and visual cues align, identifying inconsistencies that may suggest an AI-generated image.

The multimodal deep learning framework cross-references these findings with product metadata, historical transaction data, and known infringement cases to evaluate likelihood of misuse.

By combining pattern detection with contextual analysis, the technology distinguishes genuine brand imagery from generated imitations designed to bypass enforcement. When suspicious patterns are identified, listings are escalated for manual verification, ensuring both speed and accuracy.

Continuous training with verified brand data enhances model precision and reduces false positives. This layered approach allows Alibaba to effectively identify synthetic content—whether generated by AI or manually altered—while maintaining human oversight, thereby safeguarding consumer trust and protecting legitimate brands from AI-enabled counterfeiting.

What does ‘responsible AI’ mean in the context of IP enforcement? What guardrails do you have in place?

For Alibaba, “responsible AI” means applying artificial intelligence to strengthen IP enforcement in a transparent, ethical, and proportionate manner that preserves human oversight and brand expertise.

AI tools such as LLMs, image recognition, and semantic analysis are deployed to detect infringement efficiently, but decisions involving nuance or brand context remain with human experts.

Alibaba’s systems and policies include built-in guardrails to ensure proportionality, privacy protection, and explainability.

For example, automated detections trigger multi-layered review processes, escalating complex or ambiguous findings to trained human reviewers. 

Proactive systems are calibrated to minimise false positives and bias by using verified data from brand partners to train models responsibly.

Alibaba’s hybrid approach—combining deterministic rules, small machine-learning models, and human validation—ensures fairness and accountability. 

The company also limits the scope of large-scale scanning by cost and relevance, applying smaller, more precise models where appropriate.

Rightsholders play a vital role by supplying data to improve model precision, creating a virtuous feedback loop between human expertise and AI learning.

Ultimately, responsible AI in IP enforcement means using technology to enhance—not replace—human judgment, while ensuring decisions are explainable, proportionate, and respectful of all platform stakeholders.

How is Alibaba working to shape global standards in AI-powered IP enforcement?

Alibaba is taking an industry-leading role in defining responsible and effective uses of AI in intellectual property enforcement. Through continuous innovation and transparent engagement with rightsholders, regulators, and international organisations, Alibaba demonstrates how AI can be used ethically to protect IP at scale.

Its AI-driven monitoring systems integrate multimodal deep learning to analyse images, text, and transactional data in real time—setting a benchmark for industry best practices.

Alibaba’s approach emphasises the importance of human oversight, proportional enforcement, and collaboration with brand owners, helping shape emerging global norms around “responsible AI.”

By deploying its proprietary Qwen models in pilot IP applications, Alibaba also contributes to the technical evolution of AI enforcement tools while maintaining a strong compliance framework.

These real-world implementations inform broader conversations about AI governance and standards for accuracy, transparency, and fairness. Alibaba’s collaboration with brands provides valuable insights into how data-sharing and feedback loops can enhance model precision without compromising privacy or competitive integrity.

By leading these efforts, Alibaba not only raises enforcement standards within its platforms by deploying AI technologies but also influences international expectations about how AI can be applied safely and effectively to combat counterfeiting and safeguard legitimate commerce globally.

What’s your biggest concern about AI being used to recommend infringing products through social media or other channels?

Alibaba’s greatest concern about AI-driven recommendation systems promoting infringing goods is their potential to amplify harm at unprecedented scale and speed. 

When algorithms outside controlled platforms recommend counterfeit or infringing listings—whether through social media, influencer content, targeted ads or reacting to inputs by users—they can mislead consumers and erode brand trust globally. 

These AI-powered systems often lack transparency, making it difficult to trace accountability or correct misinformation once it spreads. 

Alibaba worries that bad actors could exploit these tools to automate large-scale promotion of counterfeits or misleading offers, using personalisation algorithms to target vulnerable consumer segments. 

In response, Alibaba prioritises the detection of coordinated traffic anomalies and behavioural signals that may indicate off-platform promotion. 

Its proactive monitoring infrastructure analyses inbound traffic sources and suspicious transaction clusters, allowing it to trace illicit referral patterns and remove affected listings. 

The company also collaborates with brand owners and digital platforms to share intelligence and establish stronger cross-platform safeguards. 

Ultimately, Alibaba advocates for greater transparency and shared standards governing how recommendation algorithms handle commerce content, recognising that preventing algorithmic amplification of infringement will require joint industry responsibility and global cooperation.

Looking ahead, what do you see as the next frontier in AI-powered anti-counterfeiting?

The next frontier in AI-powered anti-counterfeiting lies in fully integrated, multimodal intelligence that connects every stage of the enforcement process—from detection to evidence management and rights-holder collaboration. 

Alibaba envisions systems where advanced AI models, such as its Qwen series, combine visual, textual, and behavioural data in real time to provide a holistic understanding of potential infringements. 

Another frontier is explainable/accountable AI—ensuring that automated enforcement decisions are transparent and auditable, reinforcing trust among brands and regulators. 

However, human expertise will remain central to governance, ethics, and training data integrity. 

By advancing these technologies responsibly and collaboratively, Alibaba aims to set new standards for proactive, intelligent, and globally interoperable IP protection that keeps pace with the rapid evolution of generative AI and e-commerce innovation.