Saturday, October 3, 2026

BTS Arirang Concert Tops Ticket Scalping Reports: What You Need to Know

BTS's world tour 'Arirang' had the highest number of ticket scalping reports in the past year.

U.S. Air Force KC-135 Tanker Crash in Iraq: What Went Wrong?

A U.S. Air Force KC-135 tanker crashed in Iraq, marking the first casualties since hostilities with Iran began; rescue operations are ongoing.

Microsoft Unveils MDASH, a Multi-Agent AI Security System Built to Hunt Vulnerabilities at Scale

TechMicrosoft Unveils MDASH, a Multi-Agent AI Security System Built to Hunt Vulnerabilities at Scale
Provided by Microsoft
Provided by Microsoft

Microsoft has joined the race to develop specialized security-focused artificial intelligence (AI) models.

On Monday, Microsoft unveiled its AI-powered security analysis system called the Multi-Model Agentic Scanning Harness (MDASH).

Developed by Microsoft’s Autonomous Code Security (ACS) team, MDASH is an AI-driven security system that stands out for its ability to coordinate multiple AI agents to detect security vulnerabilities, rather than relying on a single model. More than 100 specialized AI agents work together to handle the entire process, from discovering vulnerabilities to validating them, discussing findings, and demonstrating potential exploits.

Using this system, Microsoft identified 16 new vulnerabilities across Windows networking and authentication stacks. Four of these were confirmed as critical remote code execution vulnerabilities.

MDASH has made a significant impact, achieving an impressive 88.45% success rate and securing the top spot among 1,507 tasks in the public benchmark CyberGym. This performance surpassed competing models, including Anthropic’s Claude Mythos Preview (83.1%) and OpenAI’s GPT-5.5 (81.8%).

The race to develop security-focused AI models is heating up among major tech giants. Following Anthropic’s launch of Claude Mythos in April, OpenAI responded by introducing its GPT-5.4-Cyber model. In early May, OpenAI took a step further by pre-releasing its cybersecurity-specific model, GPT-5.5-Cyber, to a select group of partners.

Microsoft believes that the discovery of AI vulnerabilities has evolved from a research endeavor to an engineering challenge. The company notes that AI-based vulnerability detection is moving beyond a competition of individual model performance, transforming into a system design-centric engineering problem.

Tae-Soo Kim, Microsoft’s Vice President of Agentic Security, stated that MDASH is helping the engineering team leverage commercial AI models to substantially improve security outcomes. Microsoft remains committed to ongoing efforts to create a safer digital world for everyone.

Check Out Our Content

Check Out Other Tags:

Most Popular Articles