Armorblox is now part of Cisco

Articles & Thought Leadership | 5 min read

From Concept to Reality: A Timeline of Armorblox's Journey with LLMs and NLU in Cybersecurity


Paige Tester
Paige Tester

GPT, LLaMa, BARD, and BERT... the gang's all here! Let's take a stroll down memory lane and recap the ways in which Armorblox has pioneered the use of Natural Language Understanding (NLU) and Large Language Models (LLMs) to guard enterprises against email-based threats and data loss.

From Concept to Reality: A Timeline of Armorblox's Journey with LLMs and NLU in Cybersecurity

Nearly six years ago, the journey of Armorblox began with a vision to harness the power of language as a revolutionary tool for safeguarding organizations against targeted email attacks and data loss.

Back then, the release of GPT 1.0 marked the early stages of a technological transformation. Fast forward to today, and the progress made in Large Language Models and machine learning has been nothing short of astonishing. As we move into the era of generative AI, we wanted to take a look back at the ways in which Armorblox has pioneered the use of LLMs and deep learning to revolutionize the way we think about protecting organizations from cyber threats.

Historical Timeline: Armorblox and its use of LLMs and NLU between 2017 and 2023

2017: “Attention Is All You Need” is published, a landmark paper by Google Research. It introduces a new neural network architecture for natural language understanding. Transformers and large pre-trained language models were developed.

2017: Armorblox was founded as the first cybersecurity company focused on using machine learning and natural language understanding (NLU) to understand the content and context of communications to detect and prevent email-based attacks.

2018: Google introduces BERT.

2018: Armorblox pioneers the use of a BERT LLM-based classifier in a production environment and trains it to detect sophisticated language-based threats.

2019: Armorblox releases its AI-powered email security solution that employs the latest advances in Natural Language Processing (NLP) and Natural Language Understanding (NLU) to analyze email content and identify phishing attempts, business email compromise (BEC) attacks, and data loss.

2019: Armorblox employs LLMs in its MLOps and Labeling Pipelines. RoBERTa and DistilBERT-based models are used for augmenting training data, consistency training of classification models, and maintaining the quality of hand-labeled training datasets.

2019: Armorblox introduces the fraud explainability feature. Customers are better able to understand the impact of fraud incidents and remediate them appropriately. RoBERTa LLM model is trained to summarize fraud email text content, and extract and highlight specific snippets and nuances from the email.

2020: T5 model released on Hugging Face.

2020: Armorblox employs the T5 model in its MLOps and Labeling Pipeline. Armorblox’s Labeling Pipeline is able to create machine-labeled training data. The T5-based fine-tuned model is used to build additional confidence in machine-labeled training data.

2020: OpenAI releases GPT-3.

2020: Armorblox’s Labeling Pipeline has built extensive training data sets for each type of fraud. Armorblox uses a new T5-based model to identify new kinds of fraud to better protect customers.

2021: T0 model released on Hugging Face.

2021: Armorblox MLOps monitors production environments for model accuracy and data drift. Armorblox MLOps incorporate T5 and T0-based models to measure model accuracy.

2021: Armorblox customers use content tags to search and categorize threats. Armorblox’s Labeling Pipeline has been extended to use the T5 model and engineered prompts to automatically extract content tags from training data examples.

2021: Armorblox continues to refine its platform and expand its customer base, helping organizations across various industries improve their email security posture. The company announces partnerships with Microsoft, Intermedia, Internet2, and Coalition, further extending the reach of its LLM and NLU capabilities.

2022: Armorblox presents at RSA Conference about the potential threats posed by generative text-based attacks and how cybercriminals may leverage generative AI tools once accessible to the public.

2022: Google introduces PaLM.

2023: Meta releases LLaMa.

2023: OpenAI releases GPT-4.

2023: Armorblox explores fine-tuning Open source Foundation LLM models on email fraud data. Armorblox is also investing in building custom Foundation Models based purely on email data.

2023: Cisco Systems announces intent to acquire Armorblox to further its AI-first Security Cloud. Cisco plans to leverage Armorblox’s trained LLMs to add generative AI to their cloud security for improved intelligence and automated capabilities.

Cisco to Acquire Armorblox

Have you heard the news? Cisco recently announced its intent to acquire Armorblox. Read more about how Cisco is furthering the AI-first security cloud.

Read the News

Experience the Armorblox Difference

Get a Demo