A hiring pause in November. A record intake the following October.

On October 13, 2026, Krafton announced that it would recruit more than 100 graduates through its 2027 recruitment-linked internship. The company described it as the largest graduate recruitment drive in its history. The openings extend beyond AI research and game engineering to corporate and commercial functions.

The contrast is sharp. During its November 4, 2025 earnings call, the publisher had said it was freezing workforce hiring outside its new intellectual-property development and deep-learning operations while pursuing an AI First transformation. That was a qualified hiring freeze, not the elimination of every position.

An internship announcement does not by itself prove that Krafton has formally withdrawn every other hiring restriction. It raises the narrower and more useful question of what kind of capability the company wants to acquire at the entry level now.

The 100-plus figure is not a promise of 100 permanent positions

The program is a recruitment-linked internship open to graduates and students expected to graduate from universities in Korea or overseas. Applications began at noon KST on October 13 and close at 11:59 a.m. KST on October 23, 2026. Position-specific criteria belong on Krafton's official recruiting website.

The announced sequence is application screening, aptitude and personality assessment, AI capability assessment, One Day Interview and final internship selection. Testing and interviews are scheduled for November and December, with results expected in January 2027.

Successful applicants are expected to start a five-month on-the-job internship in February 2027. Their conversion to regular employment is to be determined by a later assessment in July. The 100-plus target therefore refers to the internship intake, not an independently guaranteed number of permanent hires.

Six job families, one increasingly visible capability

The six families are AI research, technology, corporate or support functions, business, game development programming and game design. The company says it is looking at AI-use capability together with broad reasoning and global competence.

Producing a reliable build, managing a commercial relationship and designing a game experience are distinct kinds of work. An AI model researcher and a business analyst should not be expected to demonstrate the same technical depth. The common denominator is being able to learn and solve problems within one's own professional context using AI.

The published philosophy does not establish identical tasks, scoring weights or permitted tools across every family. Applicants should first study the requirements of their chosen role and then identify genuine examples in which AI supported work they personally understood and checked.

Cofa-Probe: a tool for looking beyond the final answer

Krafton's new hiring process includes Cofa-Probe, an AI-native hiring and assessment system it developed internally. The public question for candidates is not simply whether AI is allowed, but which parts of their problem-solving process may be evaluated.

Earlier uses offer evidence without revealing this recruitment's examination. In July 2026 Krafton and CJ Olive Young held an AI-oriented hackathon, and August reporting described Cofa-Probe's attention to how candidates define a task, guide AI, iterate and verify their work—not just what they eventually submit.

Those disclosures cannot be treated as a leaked assessment blueprint. Krafton has not established publicly that every role in this new intake receives the same exercises, duration, tool access or scoring rubric. Official subsequent instructions remain authoritative.

The evaluator had already been used outside Krafton

CJ Olive Young's July 21, 2026 newsroom announcement described a change in experienced developer recruiting: for specified positions, an AI assignment would replace the previous live coding test. The company named Krafton's Cofa-Probe as its assessment system.

Olive Young said the exercise would consider problem definition, prompt usage, revision cycles and result verification alongside the submitted product. The two companies then co-hosted a recruitment-related AI hackathon on July 30. These are concrete prior deployment contexts for the tool.

However, a retail developer's experienced-hire assignment and a game publisher's graduate internship are not the same assessment. Their subject matter, hiring goals and scoring details may differ. The meaningful connection is the attempt to observe work with AI, rather than accepting output quality as the sole signal.

How can an applicant prepare without inventing exam questions?

An applicant can review a real project in which AI played a role and reconstruct the decisions made along the way. What problem did they define? Why did they choose that approach? What was generated, what was wrong, and what evidence supported the final answer?

For a programmer, that might mean tests, debugging and code review. For a commercial or planning position, it may involve checking sources, challenging assumptions and comparing options. These examples are practical interpretations of the disclosed evaluation philosophy, not claimed examination topics.

Memorizing polished prompts may be less informative than being able to explain one's own decisions. Candidates still need to verify the actual job-specific eligibility rules, requested documents, assessment notices and the Korean-time deadline on the official site.

AI investment and hiring volumes are not the same measurement

In 2025 Krafton outlined an approximately KRW 100 billion GPU-cluster investment as part of its AI First initiative. Management framed its qualified hiring freeze around raising individual productivity and redesigning workflows, not as proof that all future hiring was unnecessary.

In its October 2026 recruitment announcement, the company emphasized the need for people able to participate in an expanding AI-led business. Neither announcement alone demonstrates how many jobs AI has replaced or whether the hiring policy for existing divisions has changed.

The operational question is more specific: how does a company combine specialists already inside the organization, newer employees fluent in AI-enabled work, and an assessment process that can distinguish meaningful capability from polished output? It takes more than a single headcount number to answer.

BANSEOG VIEW | The selection standard may change before the job title does

Between November 2025 and October 2026, Krafton publicly moved from a hiring restraint with explicit exceptions to a sizeable graduate internship targeting AI-capable newcomers. Treating this sequence merely as a hiring rebound misses the role of assessment design.

A company can declare AI fluency a priority without having proven that its chosen test fairly measures future job performance. Reviewing how candidates frame, revise and verify work may offer richer evidence, but consistency across roles and predictive validity still require evaluation. This is Banseog's interpretation, not Krafton's published validation result.

Candidates have two separate questions: Can I get to a useful answer with AI, and can I explain why the answer is dependable? Employers have a third: Can a hiring process observe those abilities without confusing tool familiarity for judgment?

BANSEOG VIEW

Banseog View — Capability selection, not simply workforce expansion

The November 2025 hiring freeze excluded new-IP and deep-learning roles; it was not a blanket, permanent ban. The October 2026 figure is a planned internship intake, not a count of confirmed permanent hires.

Previous Cofa-Probe use with CJ Olive Young and at a hackathon demonstrates an assessment approach, but not that every Krafton graduate discipline uses an identical task or rubric.

Our strategic reading is that when AI-assisted work is expected across functions, candidate evaluation must separate tool fluency, domain expertise and the capacity to validate a result. The measurement's fairness and job-performance predictive power are not established here.

SOURCES

Primary sources and references

Prepared October 14, 2026. The November 4, 2025 hiring restriction expressly excluded new-IP and deep-learning hiring; this article does not assert that all recruiting stopped or that the entire policy was officially rescinded. Krafton's October 13, 2026 announcement plans 100-plus recruitment-linked interns across six families, with permanent employment subject to later assessment. The closing time shown on employer-supplied university notices is October 23, 2026 at 11:59 a.m. KST. Descriptions of Cofa-Probe's earlier use at CJ Olive Young and an AI hackathon do not disclose the exact tasks, permitted tools or weights for all 2027 internship tracks. The assessment and talent-strategy implications are analytical interpretations rather than verified causal or predictive outcomes.