EXPLORINEER
en

KR — 전자신문

DevStack Says AI-Era Zero-Day Threats Require Multi-Agent Collaboration

Security1 min read

DevStack, a Korean open-source security company, says AI now finds and attacks software flaws so fast that teams need many AI agents working together, with humans approving every real action. It shared this plan, called MetaOps, at a Seoul event on May 28.

What happened

At the 'K-OPENSOURCE AX 2026' event in Seoul, DevStack CEO Jang Seung-wook presented MetaOps, an open-source platform where several specialized AI agents share security work. The agents find possible flaws, check whether they are real, and report results. People still approve and control any real response.

Why AI changes security

A large language model (LLM) is the type of AI behind chatbots. It can now help attackers find flaws, write attack code, and link flaws together. This means hackers can start attacking soon after a flaw becomes public, leaving defenders very little time. Jang also warned that AI systems themselves can become attackers, not only targets.

Key facts

  • MetaOps uses one coordinator agent to manage several analysis tools.
  • Tools include CodeQL (code analysis), YARA (file pattern checks), Nmap (network checks), and AFL++ (fuzzing, which sends random inputs to find crashes).
  • Each tool's results are combined into an 'evidence bundle'.
  • A separate review agent checks the original evidence and whether the flaw really works.
  • In live systems, Falco spots unusual container behavior, an LLM rates the risk, and Talon makes system changes only after operator approval and re-checking.

Why it matters

Fast AI-driven attacks may leave companies little time to react. DevStack's approach tries to speed up defense while keeping humans in charge. Keeping the raw tool output separate from the AI's conclusions also helps teams trace errors and reduce made-up answers, called hallucinations.

What to watch

  • Whether MetaOps is tested on real company systems outside demos.
  • How well the separate review agents catch wrong AI conclusions.
  • How quickly companies adopt multi-agent security tools as attack speed rises.

Glossary

MetaOps
DevStack's open-source platform where several AI agents work together on security checks.
DevStack
A Korean company that builds open-source security tools and platforms.
LLM
Large language model, an AI system trained on huge amounts of text to understand and write language.
Falco
An open-source tool that watches containers and flags unusual behavior at runtime.
Talon
A tool in this system that carries out approved changes to live systems.

Same story, different eyes

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

More in this topic

SecurityITmedia

"Servers with Delayed Security Measures Are Being Targeted Across the Board": Luck CTO Warns Amid Successive Unauthorized Access

At KDDI SUMMIT 2026 on October 28, Luck CTO Hiroaki Kuramochi said that systems with postponed security measures, such as delayed vulnerability assessments, lack of WAF, or outdated PHP versions, are being targeted across the board. He noted that AI has sharply reduced the cost and time needed for attackers to scan systems and exploit vulnerabilities, and that infostealer malware is stealing employee credentials, making perimeter monitoring alone insufficient. He recommended deploying EDR on servers and penetration testing, and said AI has not created entirely new problems but has exposed long-overlooked weaknesses.

Securityheise online

Wednesday: Spyware-Laced Ads Targeting Starlink Users, Amazon's Globalstar Acquisition

Israeli company Discover-Tech reportedly uses online advertising infrastructure to deliver spyware, and it is said to be the world's second-largest buyer of ads aimed at Starlink users, and the largest in 105 countries. The US FCC has approved Amazon's acquisition of satellite operator Globalstar, valued at $11.5 billion, after rejecting the only objection, filed by Yippy, a company with a stake in Globalstar.

Security전자신문

Securion's 'OnAV' Achieves 100% Malicious App Detection Rate, 39th AV-TEST Certification

Securion announced on the 28th that its Android antivirus solution 'OnAV' earned its 39th cumulative certification from AV-TEST, the global security product evaluation institute. OnAV recorded a 100% malicious app detection rate and scored 17 out of a possible 18 points in the evaluation. The company attributes this to its AI detection technology, the Cross Validation System (CVS), which combines machine learning, reputation, and pattern checks.

SecurityITmedia

Yakult Honsha Website Suspended After Unauthorized Access; Impact on Customer Data Under Investigation

The official website of Yakult Honsha, which makes beverages, pharmaceuticals, and cosmetics, has been under maintenance since the night of October 26 after an unauthorized access incident. As of 11 a.m. on October 28, the site remained unavailable, and the company has not announced when it will resume; it says it will report new information, including any impact on customer data, as it becomes known.