Coupang Receives South Korea's Largest-Ever Fine — Roughly $409 Million Over a Massive Data Leak

South Korean regulators have imposed a record 62.4 billion won (approximately $409 million) fine on Coupang, the country's largest e-commerce platform. The penalty stems from a massive personal data breach, making it the largest sanction in the history of South Korea's data protection regulations.

Although Coupang is a U.S.-listed company, it commands an overwhelming share of the Korean market, and this ruling has once again called into question the data stewardship responsibilities of platform operators. As AI and data-driven businesses continue to expand, the case stands as a symbol of the global trend toward stricter consumer data protection regimes.

Is LLM "Reconstruction" of AGPLv3 Code a License Violation? — Debate Heats Up on Hacker News

A Hacker News discussion — "What happens if an LLM is trained on code released under AGPLv3 and then reconstructs it?" — has drawn significant attention.

The original poster was exploring ways to prevent LLM-based reconstruction when releasing their software project under AGPLv3. Legal and ethical perspectives are clashing over where the line should be drawn between the open-source community's "freedom to use" and what amounts to de facto AI-driven code duplication.

The debate once again brings the fundamental question of how AI training data relates to licensing into sharp relief, and it could influence future legislation and license design.

AI Agents Put to Work on Test Fixes — Repairing Kong Gateway's "Flakiest Tests"

Kong, an API gateway company, has published a case study on using AI agents to fix the flakiest tests in its test suite.

The team built a workflow in which autonomous AI agents analyze test code, identify failure patterns, and generate and verify fix patches. What has drawn attention is that they automated the resolution of flaky tests — work that engineers previously had to investigate and fix by hand.

While AI agent applications in software development have centered on code generation, this is a meaningful real-world example in the quality assurance domain, specifically test stabilization.

"i1," a Fully Open Text-to-Image Model — Powerful Image Generation with a Simple Approach

"i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models" has been featured on Hugging Face's Daily Papers.

The paper presents a recipe for building strong text-to-image generation models using simple, reproducible methods rather than relying on complex architectures. Because it is released in a fully open form, a major characteristic is that the community can freely use and improve upon it.

As image generation AI becomes increasingly democratized, the growing availability of open alternatives that don't depend on commercial models is a welcome trend for the developer community.

AMD's Unified Memory Architecture Gains Traction on LocalLLaMA — New Possibilities for Local LLM Environments

AMD's Unified Memory Architecture has become a hot topic in Reddit's LocalLLaMA community.

In an LLM inference landscape that remains dominated by NVIDIA, commenters are discussing how AMD's approach could offer a new option for local execution that takes advantage of large memory capacities. In particular, Unified Memory's CPU-GPU memory sharing could prove advantageous for running large-scale models.

Users have even reported running Qwen 3.6 35B-A3B on a 32GB unified RAM setup, showing that hardware diversification is steadily lowering the barrier to running LLMs locally.

---

**Sources:**

[South Korea Fines Coupang $409 Million for Large-Scale Data Leak](https://www.bloomberg.com/news/articles/2026-06-11/south-korea-fines-coupang-409-million-for-large-scale-data-leak?srnd=phx-technology) — Bloomberg

[Ask HN: Releasing code under AGPLv3, but want to block LLM reconstruction?](https://news.ycombinator.com/item?id=48485069) — Hacker News

[How We Used Agentic AI to Fix Kong Gateway's Flakiest Tests](https://konghq.com/blog/engineering/how-we-used-agentic-ai-to-fix-kong-gateways-flakiest-tests) — Kong

[i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models](https://huggingface.co/papers/2606.11289) — Hugging Face Daily Papers

[AMD touts the unified memory architecture](https://www.reddit.com/r/LocalLLaMA/comments/1u2l25d/amd_touts_the_unified_memory_architecture/) — Reddit LocalLLaMA