Isao Takaesu (@bbr_bbq)

He is senior engineer in MBSD, involved in research related to the detection of vulnerabilities in Machine Learning (ML) systems and developing cybersecurity products. He has presented his research at hacker conferences such as Black Hat Arsenal, DEFCON DemoLabs and CODE BLUE. In recent years, he has made contributions to education as an instructor at security camps.".

Daiki Ichinose (＠mahoyaya)

He is an engineer and pentester in MBSD. He has over 15 years of work experience, and he uses his know-how to give talks at conferences such as Bsides Tokyo (2018, 2019), JAWS Days 2019, and many others. He enjoys finding vulnerabilities and loves Perl.

BLADE: An Autonomous AI Agent-Based Penetration Testing Tool – Automating the Entire Attack Chain from Attack Surface Discovery to Internal Compromise –

As cyberattacks become increasingly sophisticated and complex, the need for efficient and comprehensive penetration testing has grown significantly. This presentation introduces the design concept, core features, and a live demonstration of BLADE, a penetration testing tool powered by autonomous AI agents.

BLADE automates traditional penetration testing tasks such as privilege escalation and lateral movement. More notably, it enables the AI agent to autonomously execute the entire attack flow—from target discovery to exploitation—without human intervention. The process begins with BLADE’s built-in Attack Surface Management (ASM) module, which gathers publicly exposed servers by analyzing a target company’s name or domain. The collected data is enriched with Whois information, web content, and company-specific knowledge stored in a Retrieval-Augmented Generation (RAG) vector database. Using this information, BLADE automatically classifies whether the servers belong to the target organization.

Based on this classification, BLADE generates a custom password list by inferring likely patterns using the company’s abbreviation, founding year, and other related attributes. It then initiates SSH authentication attempts using the generated list. Once access is gained, BLADE autonomously checks the current privilege level and searches for privilege escalation vulnerabilities using tools such as LinPEAS. If a viable path is found, BLADE exploits it to achieve root-level access, often via cron-based execution or reverse shell techniques.

By automating this end-to-end process, BLADE enables fully autonomous execution of external attack surface identification and internal system compromise. Through the demonstration, this presentation highlights how autonomous AI agents can greatly streamline and enhance the efficiency of penetration testing, making it more scalable and adaptable to evolving threats.