KT seeks five years of penetration-testing experience and the ability to build a pentest agent

KT opened an October 26–November 1, 2026 recruitment round for a permanent Red Team role based in Pangyo, South Korea. It requires at least five years of experience in APT and web or application penetration testing, together with knowledge of web, API, system and network assessment, attacker tactics and MITRE ATT&CK scenarios.

The same posting calls for LLM-assisted security analysis, source-code and vulnerability assessment, AI-enabled security-diagnostics automation, and Pentest Agent development. The company is looking for expertise both in performing penetration tests and building tools that may carry out parts of the work.

This is not KT's first AI pentesting vacancy. Its September 15–27 recruitment already listed similar AI agent development skills. The October listing clarifies the Red Team and AI Pentesting role, but does not prove that KT created a new division, expanded headcount or failed to fill the earlier vacancy.

One job description combines three different skills

The first is hands-on penetration testing: identifying a plausible weakness and checking whether it creates an exploitable risk in a customer's actual environment. Scanner alerts alone do not establish meaningful impact.

The second is attack-path reasoning. Using knowledge such as the MITRE ATT&CK framework, a specialist designs controlled scenarios to test how several actions might interact with defensive systems, rather than treating each finding as an isolated defect.

The third is implementing repeatable analysis and testing tasks in AI-driven software. KT lists all three within a single role. It has not disclosed how much working time would be allocated to each function or whether any specific agent has already reached full production.

Penetration testing was automated long before LLMs arrived

Vulnerability scanners, scripted checks and testing frameworks have been used for years. AI did not invent automated security assessment. The more specific possibility is to apply language models to interpret intermediate results, help select approved follow-up checks and organize evidence for the human analyst.

An AI agent may suggest or execute additional tests under explicitly authorized conditions. That does not mean it should make unrestricted changes to a customer's systems or decide unilaterally that an exploit is safe to run.

Increasing autonomy can increase the importance of scope control, permitted tools, human review, stop conditions and reproducible logs. Automating portions of expert judgment is different from transferring final accountability to the software.

SK Shieldus describes the conversion of expert know-how even more directly

SK Shieldus advertised a permanent PEN-AI service development and diagnostics role starting October 22, with at least three years of vulnerability-assessment experience and a Pangyo work location. Its stated duties include web vulnerability analysis, external attack-surface mapping and assessment of linked attack scenarios.

The posting calls for AI-agent-based penetration-testing automation and automated attack-scenario generation, execution and results analysis. Particularly revealing is the requirement to translate security specialists' diagnostic know-how into AI and automated logic.

There is no evidence of a joint KT–SK Shieldus product or an identical service model. But the two independent job advertisements show a shared demand for professionals who can combine practical offensive-security expertise with AI automation engineering.

Can a specialist's security judgment become reusable company software?

Experienced penetration testers must distinguish real risks from false alarms by interpreting configurations, user permissions, application context and multiple pieces of evidence. A high volume of scanner findings is not equivalent to a high-quality assessment.

Some repeatable rules and procedures may be encoded in software: specialists define the tests, approved agents execute selected checks and organize evidence, and qualified humans validate severity, reproducibility and remediation.

The model also creates engineering, maintenance and assurance costs. Neither company has disclosed that agent-based testing is already more accurate or less expensive. Reusable value depends on whether the procedure works across different authorized environments without sacrificing trust.

Security for AI is not the same job as using AI for security

AI Security can involve evaluating AI systems themselves, including LLM applications, retrieval-augmented generation, agents and their potential abuse. AI Pentesting in this case means using AI as part of the toolset to inspect conventional applications, networks and systems.

KT's vacancy and the SK Shieldus PEN-AI development posting center on the second type, though operating AI tools securely still matters. A separate SK Shieldus AI security research vacancy explicitly addresses LLM, RAG, MCP and AI-agent threats, offering a useful contrast.

A recruiter who treats every opening containing the phrase 'AI security' as interchangeable risks sourcing the wrong candidates. The target being assessed, the role played by AI and the responsibility for the result must be defined.

Who decides whether an agent's finding is a genuine vulnerability?

An AI system may report many suspected weaknesses but create more manual verification work if a large share cannot be reproduced. Conversely, a smaller set of high-impact, validated findings may be worth considerably more.

An evaluation could separate true finding precision, meaningful assessment coverage, total expert time including validation, compliance with approved scope and the ability to reproduce recorded outcomes. Remediation and customer impact would require additional evidence.

These are Banseog's suggested evaluation dimensions, not published KPIs for KT or SK Shieldus. More automation steps and more generated reports do not by themselves prove better customer security or commercial returns.

Do these vacancies prove that security jobs will disappear?

No. KT explicitly asks for at least five years of relevant practical experience; SK Shieldus requires at least three. Neither vacancy demonstrates industry-wide job losses or full replacement of skilled penetration testers.

Instead, the combination of capabilities may change: an experienced offensive-security analyst may need software and AI automation skills, while an agent engineer needs enough security judgment to know which outcomes matter and which actions are safe.

For hiring teams, the critical evidence is not merely a list of AI tools used. It is the ability to show what was automated, how false findings were assessed, how permissions and boundaries were enforced and how security professionals retained oversight.

What must candidates verify before applying?

KT's role is permanent, based in Pangyo, and open through November 1 according to its October 26 posting. It specifies a completed four-year university degree, relevant penetration-testing experience of five years or more, and eligibility relating to military service for male applicants and international travel. Experience with ATT&CK, AI-based diagnostics, hackathons and bug bounties may be considered.

SK Shieldus PEN-AI service development is a separate permanent vacancy, also based in Pangyo, with at least three years' diagnostic experience and a stated November 15 application deadline. Applications should be completed through the relevant employer's official recruiting channel.

KT's deadline date is consistent across the reviewed notices, but third-party sites disagree on the exact closing hour. Candidates should check the official portal, not infer a time from this article. Neither employer's final hiring headcount or exact project assignment has been publicly verified.

Recruitment demand is evidence of a direction, not an achieved operating result

The job postings show employers asking for security judgment and AI automation engineering together. They do not prove KT moved all penetration tests to AI agents, that September hiring failed, or that the two companies are collaborating.

The public notices also do not establish actual service revenue, precision gains, lower labor costs or the percentage of an assessment that is already performed autonomously. These remain separate questions to be tested with operating data.

For a strategic reader, the shift is not a new buzzword in job titles: it is a potential move from expertise residing only in individuals to expertise that can be encoded, reused, verified and controlled inside an organization.

BANSEOG VIEW

BANSEOG VIEW | Encoding expert judgment increases the need to validate it

Both KT and SK Shieldus are hiring people who can bridge hands-on offensive security, meaningful attack-path assessment and AI automation engineering. This is different from simply wanting staff who have tried a chatbot.

If repeatable expert procedures are encoded in software, portions of diagnostic work may become reusable organizational capability. That also raises the stakes of scope control, false-positive review, reproducibility and accountability for findings.

Hiring teams should evaluate what candidates actually automated, the accuracy and safety of their systems, and how specialists validate results. These vacancies do not yet prove realized cost savings, better security outcomes or job displacement.

SOURCES

Primary sources and references

Prepared Oct 28, 2026. KT's employer-posted Oct 26–Nov 1 Red Team vacancy requires five years of related experience, Pangyo work and AI Pentesting development. SK Shieldus's Oct 22–Nov 15 PEN-AI listing requires three years of diagnostics experience. KT's September role already mentioned agent development; October is not its first known AI penetration-testing hiring. Third-party listings differ on the precise KT cutoff hour, so the official portal takes precedence. Neither listing establishes deployed autonomy, business impact, job reductions or staff headcount. Distinction between AI Security and AI-assisted penetration testing is maintained; the suggested KPIs are conditional Banseog analysis.