김하영 is a journalist at META-X.
PUBG Asia Stars 2026 began with stream-sniping allegations against two Vietnamese players, followed by eight Korean players withdrawing and the final day's cancellation. Permanent bans followed, while the controversy raised broader questions about fairness, rules, and tournament operations.
Generative AI has largely run in massive data centers, but local AI is bringing some of that computing back to personal computers. With enough GPU power and memory, a PC can run models and serve them to other devices, giving older desktops and workstations a new role as small AI servers.
EvilTokens used AI to analyze stolen emails and identify targets, while Microsoft’s MDASH uses AI to find vulnerabilities in code. As AI speeds up both attack and defense, security teams face a new challenge: deciding what to fix first and responding before attackers can exploit it.
Hancom is expanding AI services that turn public documents into searchable, usable data. HWPX is now an open XML-based format, while Japan’s Ichitaro and China’s WPS show similar local ecosystems. In the AI era, the key issue is keeping the freedom to manage and use domestic knowledge.
Riot’s live-service success with ‘League of Legends’ and ‘VALORANT’ contrasts with Marc Merrill’s view that “not every game needs to be infinite.” ‘2XKO’ and growing time pressure suggest that each game may need its own lifespan and operating model.
Meta is reportedly developing “Luna,” smart glasses without a camera or display. While cameras became the eyes of AI, they also raised privacy concerns for people nearby. By focusing on AI and voice, Luna points to smart glasses designed to blend more naturally into everyday life.
Anthropic is operating a wet lab, bringing Claude closer to real-world life sciences research. The move suggests AI companies may expand beyond model performance toward creating and learning from new experimental data.
Shift Up’s generative-AI music video ‘Wanna be in LOVE’ drew criticism less for using AI than for awkward faces, unnatural motion, and inconsistent visuals. Fans also felt the experiment weakened the studio’s established visual identity, raising a simple question: what did AI actually add?
Naver Cloud and LG CNS are developing separate offensive and defensive AI models for cybersecurity. HyperCLOVA X will focus on defense, while an EXAONE-based model strengthens attack capabilities, aiming to build controllable AI for critical infrastructure.
Berlin artist Simon Weckert’s Digital Camouflage disrupts person detection by exploiting how object-detection models read human figures. Following CV Dazzle, adversarial patches, and the Adversarial T-shirt, it explores how far fashion can go in confusing machine vision.
Anthropic’s clash with the U.S. Defense Department centers on who sets the limits of military AI use. The Pentagon demanded “all lawful uses,” while Anthropic kept restrictions on domestic surveillance and fully autonomous weapons. The court blocked the sanctions, leaving the core issue unresolved.
Jensen Huang said AGI may already be here, but what matters is what AI can do. ARC-AGI-3 showed the same model can perform very differently depending on its system. inZOI and MIMESIS suggest that in the AGI era, where and how intelligence is deployed may matter more than intelligence itself.
Instagram’s First Draft turns multiple clips into a rough edit in-app. The feature is not new, but placing it within Instagram cuts the need for separate editing tools, lowers barriers for beginners, and makes planning, selection, and creative judgment more important as editing becomes automated.
Meta’s Pocket lets users create small games with natural-language prompts, share them through a feed, and remix what others have made. Remixing is not new, but Pocket links AI creation, discovery, and re-creation in one flow, potentially adding remixability to how UGC games are valued.
The original Rainbow Six emphasized planning over gunplay, from choosing operators and routes to timing assaults with go-codes. Over time, the series shifted toward real-time decisions and multiplayer coordination. Rainbow Six Tactics asks whether that planning can still feel compelling in 2027.
Since August 2026, Claude has added statistical watermarks to generated text and C2PA provenance data to supported files. The watermark leaves patterns in word choice rather than hidden characters, but editing can weaken detection. Users still cannot directly verify whether the signal remains.
The 2023 AI pause failed without shared risk rules or verification. In 2026, Astra neared the Critical cyber threshold, prompting OpenAI to pause some training and evaluation. Safety standards began to shape development, though companies still largely set and verify the rules.
Netflix is licensing existing YouTube creator videos on a non-exclusive basis, creating a second distribution window. YouTube, in turn, is considering extra payments for early-release or limited exclusivity deals. For now, this competition is concentrated among a small group of top creators.
For 35 years, 'Super Robot Wars' has preserved its crossover core while changing how it is made. Its move to Unity, shifts in key roles, and organizational changes show the challenge of sustaining a long-running IP, while DLC and remasters help keep the series active between new releases.
OpenAI used Astra to produce new solutions and advances on 10 open problems in math and theoretical computer science. Lean 4 verified logical validity, but debates remained over novelty, prior work, and credit. As AI enters research, humans may focus on judging value, originality, and attribution.