Researchers from the Indian Institute of Science and the Defence Research and Development Organisation have unveiled HybridNet, a multimodal framework designed to detect fake news by identifying mismatches between text and images.

The system represents a significant step in the fight against misinformation, particularly in an era where manipulated content spreads rapidly across digital platforms.

HybridNet’s core approach is to treat open-source Vision Language Models as reasoning tools rather than conventional classifiers.

This allows the framework to move beyond simple labelling and instead perform deeper logical checks on the consistency of multimodal content. By leveraging reasoning capabilities, the system can provide more reliable authenticity assessments.

The framework runs a three-stage consistency check. First, it examines image-text alignment to determine whether the visual content matches the accompanying narrative.

Second, it performs image-image comparisons to detect duplication, manipulation, or inconsistencies across visual datasets.

Third, it conducts text-text checks to identify contradictions or semantic mismatches within the written content itself. This layered approach ensures that HybridNet can capture subtle discrepancies that single-modality systems often miss.

HybridNet is designed to be lightweight. It distils observations into a compact classifier that generates clear, human-readable explanations alongside authenticity scores. This transparency is critical, as it allows analysts and end-users to understand why a particular piece of content has been flagged as suspicious, rather than relying on opaque machine outputs.

The system also incorporates active learning to reduce the burden of data annotation. By intelligently selecting which samples require labelling, HybridNet maintains high accuracy while using labels for less than half of the training data. This efficiency makes the framework more practical for large-scale deployment, where manual annotation is often a bottleneck.

Beyond its technical design, HybridNet reflects a broader push by IISc and DRDO to develop indigenous AI solutions for national security and information integrity. The framework is particularly relevant in the context of defence and strategic communication, where misinformation campaigns can have serious consequences.

By combining multimodal reasoning with efficient learning strategies, HybridNet positions itself as a robust tool against adversarial information operations.

This initiative also aligns with India’s wider focus on artificial intelligence for defence applications. DRDO has previously emphasised the importance of trusted indigenous AI systems, particularly in areas such as cyber defence and threat intelligence.

HybridNet complements these efforts by addressing the information warfare dimension, ensuring that fake news detection is not only accurate but also explainable and resource-efficient.

The project demonstrates how academic institutions and defence organisations can collaborate to tackle pressing technological challenges. IISc’s expertise in advanced AI research and DRDO’s focus on national security applications create a synergy that strengthens India’s capability to counter misinformation.

HybridNet is therefore not just a technical achievement but also a strategic milestone in safeguarding information ecosystems.

Agencies