Semantic AI: Contextual email threat detection

Semantic AI by Libraesva: the contextual AI email security engine that reads between the lines

Libraesva Semantic AI

In response to the increasing sophistication of email-based attacks, Libraesva has developed an essential new layer of protection that goes way beyond simply looking at content. Semantic AI is the contextual AI engine that detects subtle anomalies, unexpected behavioral patterns, and malicious intent, even when a threat is buried in complex, legitimate-looking email threads.

This is game-changing technology

Libraesva’s AI-driven Adaptive Trust Engine (ATE) intelligently assesses sender relationships, communication patterns, and threat likelihoods to protect businesses around the world from email security threats.

Now, running in parallel with ATE, Semantic AI is powered by a lightweight and highly specialized small language model (SLM). This discriminative AI engine analyses the meaning, intent, and context of every email to expose even the most convincing attacks. Its zero-entropy processing is critical in security operations where repeatability and trust are non-negotiable — a decisive advantage over probabilistic large language models.

AI Versus AI Is the Next Battleground for Email Security.

Paolo Frizzi, AI Business

Semantic AI is faster, lighter, and smarter

Unlike bloated cloud-based AI gimmicks, Semantic AI is lightweight, fast, and designed to run directly on your existing gateway hardware. Developed entirely in-house by Libraesva, Semantic AI introduces no risky dependencies on third-party APIs or cloud-based AI platforms. For you, that means no GPUs, no offloading to the cloud, no new vulnerabilities.

Instead, Semantic AI delivers what truly matters in email security: clear classification, contextual understanding, and consistent performance, all running invisibly in parallel to your other defenses, silently keeping your inbox safe.

Benefits of running the contextual AI email security engine

  • Clarity

    Semantic AI doesn’t create, it classifies. It uses the same neural network and tensor-based technology as generative AI to deliver precise threat classification, with no hallucinations and no synthetic outputs.

  • Contextual irregularities are identified

    Why would your CFO reference a payment link to a vendor that your company’s never dealt with? Instead of looking for keywords or behavioral signatures, Semantic AI picks up on the inconsistencies in logic or context that rule-based engines can miss.

  • Smarter and faster operation

    This isn’t a generic trillion-parameter language model trained on internet noise. Our engineers have created an agile 100M-parameter model trained on thousands of curated real-world threat examples. With latency of less than a second, Semantic AI runs on standard CPU hardware: no delays, no GPUs, no need to offload to the cloud.

  • Day 0 detection

    During internal testing, Semantic AI detected fake ADSL subscription scams on Day 0, before any signature updates or pattern propagation. It proactively interprets anomalies in communication, catching threats early and decisively.

  • Parallel protection

    The Libraesva Semantic AI engine adds a deeper layer of scrutiny to your existing protection without adding complexity.

  • Discriminative intelligence

    Unlike generative models, which introduce random variability into their results, Semantic AI is discriminative. The same email will always yield the same output, for the consistent, reliable classification that’s essential for security reliability.

Discriminative AI learns to distinguish between different types of data, making it ideal for tasks requiring sorting data into categories. For example, it can identify whether an email is spam, recognize objects in an image, or diagnose diseases from medical scans.

TechRepublic, October 2024

  • Weekly retraining with high-frequency updates

    Libraesva’s agile approach ensures rapid responsiveness to emerging threats without burdening your system with full retraining cycles.

    Semantic AI’s language model is initially trained on the semantics of 100 languages for broad foundational semantic understanding. We then fine-tune this base model for email classification using curated threat examples – this takes just a few hours, and updated models can be deployed within minutes.

Interested in learning more?

Download the datasheet here!