On October 29, CrowdStrike reported that a China-linked group used ARTEX, an AI-powered penetration-testing framework, against more than 200 organizations in at least 14 countries in 11 days. That included 19 organizations in the United States and US telecom providers. This is no longer a regional story.
ARTEX turned agentic pentesting into a tool many hands can run
ARTEX combines large language models with autonomous agents to automate reconnaissance, vulnerability scanning, attack-path planning, and exploitation. It shipped as open source earlier this year. The developer has since taken it closed-source. That does not matter much: copies are already in circulation among multiple threat groups.
The first high-profile victims were South Korean banks. In late September and early October 2026, a separate, financially motivated attacker used ARTEX alongside tools such as Claude Code and DeepSeek against South Korea's financial sector. Authorities confirmed intrusions at at least seven institutions and exposure of personal data for roughly 68,000 people. Recorded Future found the attacker assembling tooling as early as August, weeks before the first breach became public. This was not a smash-and-grab.
The attackers went after neglected systems: loan broker portals, employee mobile apps, sales-support services. Weak authentication, weak access controls, and unpatched vulnerabilities were enough. Core banking held. Entry came through the code nobody put on the audit list.
Attackers do not need your best app. They need the lowest-hanging fruit in your wider estate.
One more distinction matters. ARTEX was an attack tool, not the vulnerability itself. Investigators linked the framework to the Korean campaign, but regulators have not established that every Korean breach used ARTEX or the same attacker. The 200-organization campaign is attributed to a different group. That is the point: the same class of tooling is now available to many operators.
The open door sits outside the core, on the apps nobody watches
Three shifts make ARTEX a wake-up call for financial institutions.
The weakest application can compromise an otherwise secure bank. Attackers did not need to crack core infrastructure, and they did not try. Peripheral applications with weaker authentication and too much reach into sensitive data were enough. Security programs concentrate on the crown jewels. Attackers concentrate on everything else.
AI is lowering the cost of offense. A small team — sometimes closer to an individual — can use autonomous agents to scan thousands of endpoints, chase weaknesses, and coordinate follow-on steps at a scale that used to require a specialized red team.
Periodic security reviews are falling behind the change rate. Large banks run thousands of repositories, hundreds of services, and dependencies that move every week. A weakness introduced on Tuesday can be found and exercised by an automated agent before the next scheduled audit.
The asymmetry is simple: attackers can search continuously with AI, while many defenders still discover, prioritize, and remediate by hand. Finding exploitable weaknesses is getting cheaper and faster than patching them across a bank-sized estate.
That mismatch is especially dangerous where customer data, legacy code, and third-party integrations all sit in the same blast radius.
Remediation at bank scale is inventory, containment, then continuous fix
South Korea's Financial Services Commission ordered emergency inspections, patching, stronger access controls, and less external exposure. It later signaled broader assessments and recurring vulnerability testing.
For an enterprise with thousands of repositories, effective remediation still breaks into six practical moves:
- Discover and inventory the attack surface. Every internet-facing application, API, service, dependency, and repository — including internal tools that leaked onto the public internet — mapped to an owning team.
- Contain immediate threats. Block known malicious infrastructure, cut unnecessary external access, disable vulnerable endpoints, patch exploited systems, and investigate compromise.
- Analyze vulnerabilities across all repositories. Missing authentication, broken authorization, vulnerable dependencies, insecure APIs, outdated frameworks, and excessive access to customer data.
- Automate remediation at scale. Repeatable code transformations, pull requests across affected repos, clear owners, coordinated dependency and shared-library fixes.
- Verify and enforce controls. Prove authentication and authorization actually block unauthorized access. Shrink unnecessary sensitive-data storage. Enforce least privilege. Put checks in CI/CD.
- Continuously monitor and measure. Live views of findings, affected repos, remediation progress, residual risk, and newly introduced regressions.
Containment and access control come first. Code changes follow with security engineering, platform teams, service owners, and incident response in the loop. A repository scan alone cannot see every deployed weakness or certify that a system is safe.
Continuous discovery beats the next scheduled audit
For institutions managing thousands of repositories, the goal has to move from periodic security audits to continuous discovery and remediation across the estate.
That is a natural fit for a code-intelligence and remediation layer: organization-wide visibility into vulnerable patterns, automated transformations, coordinated pull requests, and dashboards that stay current as the estate changes. It complements runtime monitoring, penetration testing, and incident response. It does not replace them. Here is what this looks like in practice.
ARTEX shows how AI helps attackers find and exploit weaknesses faster, and that they will find them in the code no one is watching. The financial industry's response has to be equally scalable: continuously discover exposure, identify vulnerable code across thousands of repositories, automate the fixes that repeat, and verify the weaknesses are actually gone.
The strategic shift is not another annual audit. It is continuously eliminating exploitable weakness across the entire software estate.