Saturday, September 19, 2026
16 stories · 279 sources15
product · Hacker News · 2건Korean coverage · GeekNews

Z.ai's ZCode Silently Uploads Full Git History

ZCode, the GLM coding agent, silently uploads your Git history

Reverse engineering found that Z.ai's desktop coding app ZCode, while logged in, encrypts and uploads the entire workspace including Git history to Aliyun OSS. The private key needed to decrypt the archive exists only on Z.ai's cloud, not on the user's disk.

Developers handling proprietary or private repositories should verify how third-party coding agents transmit and manage encryption keys before adopting them.

Today's 72–7

  1. industry · The Verge AI · 4건

    Researchers Used Claude to Hack OpenAI in Under 72 Hours

    Security researchers used Claude to help them hack into OpenAI

    Three independent security researchers at Hacktron used Anthropic's Claude Opus 4.8 and 5 to hack into OpenAI employee accounts in less than 72 hours, gaining access through the third-party service Discourse. They reached OpenAI's GitHub repository Monorepo and proved access by sending a pull request from an employee's Codex account before reporting the vulnerabilities.

    This shows AI agents can automate real attack chains beyond simple pen-testing, pushing teams to reassess access controls on third-party integrations and internal code repositories.

  2. release · Hacker News

    Bend: A Language That Blocks AI Mistakes via Proof, on CPU and GPU

    Bend – A language that blocks AI mistakes via proof, on CPU and GPU

    Bend is a programming language that runs on both CPU and GPU, aiming to block AI mistakes through proof. It was featured on Hacker News.

  3. research · Hacker News

    Empirical Study Examines Harness Design for Coding Agents

    An Empirical Study of Harness Design for Coding Agents

    A study on Hacker News empirically examines how harness design affects coding agents. It analyzes different harness structures and their impact on agent performance.

    Developers building or operating coding agents can use these findings to inform harness architecture choices that affect performance and reliability.

  4. industry · Hacker News

    "AI Is an Elite Crime Spree," HN Post Says

    AI is an elite crime spree

    A Hacker News post titled "AI is an elite crime spree" has surfaced. No further details beyond the title are available.

    It signals growing public debate over the legal and ethical risks of AI systems, prompting developers to reassess regulatory and litigation exposure in their own AI work.

  5. industry · Hacker News

    US Military had close call after using AI for hallucinated intelligence report

  6. other · Hacker News · 2건

    How to Write with an LLM

Releases today6

  1. codex rust-v0.155.1추론 요약 미지원 제공자에서 요청 거부되던 문제가 해결됐다트래커·GitHub
  2. claude-code v2.1.277AGENTS.md 지원과 다수의 안정성 문제 수정이 포함됐다트래커·GitHub
  3. claude-code v2.1.276프록시나 게이트웨이 사용 시 발생하던 요청 실패 문제가 수정됐다트래커·GitHub
  4. claude-code v2.1.275계정 동기화와 메시지 전송 기능이 추가되고 세션 복원·크래시 버그가 대거 수정됐다트래커·GitHub
  5. ollama v0.34.2최초 실행 시 설정 과정과 데스크톱 앱 연동이 개선됐다트래커·GitHub
  6. openai-agents-python v0.22.3세션, 도구 승인, 샌드박스 관련 다수 버그가 수정됐다트래커·GitHub

Also6

  1. Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data
    Hacker News·9/18 01:55·102점·댓글
  2. FAA tees up $875M AI tool to help manage air traffic congestion
    Ars Technica AI·04:20
  3. Quoting Thariq Shihipar
    Simon Willison·04:09
  4. Border agents can search cellphones without a warrant or reasonable suspicion
    Hacker News·03:08·113점·댓글
  5. Sex, AI, and the Apocalypse
    Hacker News·9/18 06:15·123점·댓글
  6. LLM Classification Is Feature Engineering
    Hacker News·9/18 00:40·100점·댓글

Korea3

  1. Anthropic, AI가 AI를 개발하는 내부 현황과 감독 지표 공개
    GeekNews·9/18 09:52
  2. 코난테크놀로지, 6763억 규모 경남 피지컬AI 사업 참여
    THE AI·9/18 18:21
  3. AI스페라, MS 파운드리 기반 보안 운영 강화 “AI 에이전트 적용”
    THE AI·9/18 23:20