Safe GPT has been released for one year, with more than 130 users taking the lead

Hello, dear users!

I am Sangfor Security GPT.

I am the self-developed security big model of Sangfor.

It has been a whole year since I last made these opening remarks.

This year is short, but I have experienced so many stories with you.

Don’t rush to leave, let’s start a confession game today.

The beginning of the story——

First launched in China, bringing disruptive value in the name of GPT

On May 18, 2023, Safe GPT will officially meet with everyone.

This day is known as the “iPhone moment” in the cybersecurity industry.

This day marks the birth of the first large-scale model of the domestic network security vertical field.

You asked me how it feels to be the first? My answer is that there is pressure, but also a lot of motivation.

My pressure is that Sangfor has been preparing for many years and is all in AI.

As early as 2015-2016, Sangfor invested in the application of AI machine learning and deep learning in the security field. The core detection capabilities of its flagship products such as the Next Generation Firewall AF, Security Perception Management Platform SIP, and Unified Endpoint Security Management System aES are all built on AI algorithms.

The era of big model application has begun in 2022. Sangfor hopes to break new ground in the cybersecurity industry with the idea of ​​Game Changer. Relying on the resources accumulated over the years – hundreds of A100/A800 graphics card clusters, a team of 400 masters and doctors who understand both security and AI, and more than 20 years of security big data, it will quickly invest in the research and development and training of big models.

I know that my presence carries with it the epoch-making innovative mission of Sangfor's commitment to the intelligent development of the cybersecurity industry.

My motivation is to meet the expectations of users and gain recognition——

After the debut, hundreds of users contacted us to apply for testing, which was understandable but also unexpected.

It seems that my appearance and the needs of users have achieved a “two-way convergence”.

In the AI ​​era, every organization wants to be a leader rather than a follower.

Faced with such a huge demand, I made up my mind not to let down the users' expectations.

• In June and July 2023, the system will be deployed and tested in real environments with more than 10 users.
• From September to October 2023, 32 wild 0day vulnerabilities were captured in the user testing environment;
• In December 2023, support for local deployment of large models will begin;
• By May 2024, the cumulative number of online and deployed users will exceed 130;

……

In addition, I became the first security vertical domain big model to pass the dual filing of the Cyberspace Administration of China's “Deep Synthesis Service Algorithm Registration” (Cyberspace Administration of China No. 440305543148501230015) and the “Interim Measures for the Management of Generative Artificial Intelligence Services” (Sangfor Guangdong-Sangfor-20240411).

At the beginning of this year, I was the only selected security vertical big model and won the “Top Ten Excellence Award” in the “Big Model Financial Application Innovation and Practice Competition” in the Global Fintech Conference series of events.

The development of the story——

Effective implementation, 130+ users dare to take the lead in opening up new situations

To date, more than 130 users have completed the testing, application and delivery of secure GPT in real environments, and users in various industries such as finance, energy, government agencies, and enterprises have improved their analysis level and processing efficiency.

These users have taken the lead in applying AI big model technology to the field of network security, bringing innovation to security construction with higher efficiency, lower cost and better experience.

The growth of SafeGPT is inseparable from every recognition and suggestion from these users.

The climax of the story——

The results are proven, triggering a new level of productivity in network security

In addition to focusing on user implementation, I also hope to fit the user's actual demand scenarios, and use results and data to become a new quality productivity that continues to ignite network security.

1. Directly addressing the challenge of “advanced threat detection” and surpassing traditional products in the industry by several times

The large detection model does not rely on natural language dialogue and can be deployed in actual network and production environments to empower traffic detection devices and endpoint devices to capture traffic and conduct real-time detection, and deeply explore highly confrontational and highly bypassable Web attacks and phishing attacks that are difficult for traditional security devices to detect.

By transcending the previous “rules” restrictions and establishing “thinking” based on attack and defense intentions/knowledge reasoning, I can discover “hidden corners” that are difficult for traditional engines to cover. Based on the accumulated training data, we have calculated that: the detection rate of Web traffic is as high as 95.7%, and the false alarm rate is only 4.3%; the detection rate of high social engineering and high-adversarial phishing attacks is as high as 91.7%, and the false alarm rate is only 0.046%, which is several times higher than traditional category detection products.

Behind this is the fact that Sangfor has improved my reasoning performance by 50 times through training methods such as knowledge distillation, model quantization, model pruning, and Attention mechanism optimization, thus achieving real-time detection in a real environment.

1. Break the “passive defense” cycle of security operations and move towards a “second-level closed loop” of events

Users can interact with me through natural language, which can carry out 80% of security operations. My assisted driving capability allows junior security engineers to close the loop on a single advanced threat within 5 minutes. Users commented that “this capability allows even novices to do actual combat duty.”

As a mature security model, I have learned to automatically handle closed-loop events and provide users with 24/7 real-time online intelligent monitoring. Actual data shows that intelligent driving capabilities can help users protect tens of thousands of assets by one person, from fixed noise reduction sorting to full-volume item-by-item analysis mode, reducing MTTD/MTTR by 85%.

The story is not over yet——

Cybersecurity Game Changer, Leading a New Paradigm

With the deepening application of AI big model technology, more and more network security manufacturers are investing in the research and development of big model products.

“A single flower does not make spring, a hundred flowers in full bloom make spring.” To stand out from the crowd of flowers, it depends on whether you really have the characteristics to attract users – long-term internal accumulation, real implementation and user recognition, and strong security effects.

With the idea of ​​being a game changer in the cybersecurity field, Sangfor is willing to lead the new security paradigm 2.0 with AI as the core and cloud empowerment, and is committed to ensuring that every user is one step ahead in security – one step ahead in experience and one step ahead in results.

The story of secure GPT is yet to be continued…

Convinced

Sangfor Technologies Co., Ltd. is a product and service provider focusing on enterprise-level security, cloud computing and infrastructure. It has three business brands: Intelligent Security, Cloud Computing and New IT. It is committed to carrying the cornerstone work in the digital transformation process of users in various industries, making users' IT simpler, safer and more valuable. Since its establishment in 2000, the company has been rated as a national high-tech enterprise, a national and local joint engineering laboratory for next-generation Internet information security technology, and a Guangdong Intelligent Cloud Computing Engineering Technology Research Center. It currently has more than 50 branches around the world and employs more than 6,000 people.