2026-06-29

15 stories suggest →

Top stories

  1. GLM 5.2 tops Semgrep's IDOR cyber benchmark, beating bare Claude Code Open-weight GLM 5.2 scored 39 percent F1 on IDOR detection versus Claude Code's 32 percent, at 0.17 USD per vulnerability found. discussion AI
  2. OpenAI Codex still lacks a way to exclude sensitive files from the model Codex lacks a .codexignore mechanism to exclude secrets and credentials from being read and sent to the model, open since 2025-08-28. discussion Agentic coding
  3. Developer uses Claude Code to get a second opinion on an MRI report A developer used Claude Code to cross-reference an MRI report against clinical guidelines to question a treatment recommendation. discussion Agentic coding
  4. Librepods reaches the front page with AirPods reverse-engineered for non-Apple devices Librepods, a GPL-3.0 project, reverse-engineered AirPods control features to expose battery, noise control, and gestures on Android and Linux. discussion Dev tools

ML research

  1. Paper argues AI-agent risk should be measured at the repository level A preprint argues that evaluating coding agents one at a time misses ecosystem harm when many agents commit to shared repositories concurrently. developing
  2. ToolPrivacyBench audits whether tool-using agents leak private data to the wrong tools 2,150-case benchmark found tool-using agents completed tasks while inappropriately disclosing private data through intermediate tool calls. developing

Agentic coding

  1. "Tokenmaxxing" framed as returning on compounding-correctness economics An opinion post argues that early agent tokenmaxxing is replaced by compounding correctness, where more tokens tend to yield better outcomes. discussion

Apple platforms

  1. Reverse-engineering Apple's Sparse Image Format (ASIF) Erik Schamper reverse-engineered ASIF, Apple's sparse virtual-disk format for macOS virtual machines, and released an open-source reader. discussion

Infrastructure

  1. China's LineShine debuts at TOP500 number 1, first China system to lead since 2017 LineShine, running custom domestic processors and interconnect, became the first Chinese system to lead TOP500 since Sunway TaihuLight in 2017. confirmed

Engineering posts

  1. Cloudflare traces an intermittent image-truncation bug to a discarded Poll::Pending in hyper Cloudflare traced a six-week image-truncation bug to a discarded Poll::Pending in hyper's HTTP/1 state machine, a four-line fix in dispatch.rs. confirmed
  2. HackerRank open-sourced its resume-scoring agent; analysis finds the scores non-deterministic One resume scored 66 to 99 out of 120 through HackerRank's open-source resume-scoring agent, revealing unreliable evaluative judgment. discussion

Markets and companies

  1. Samsung, SK Hynix, and Micron hit with US antitrust class action over DRAM prices A class action alleges Samsung, SK Hynix, and Micron coordinated DRAM pricing since 2022, with prices rising approximately 700 percent. confirmed

Hacker News

  1. Essay argues age verification is a precursor to speech attribution tops the front page An opinion essay arguing mandatory age verification precedes persistent identity attribution of online speech topped the front page. discussion
  2. Stanford data page charts memory prices from 1960 to 2026 A Stanford data page plotting per-byte prices for DRAM, flash, and disk memory from 1960 to 2026 reached the front page with 167 points. discussion
  3. GLM 5.2 benchmark thread, methodology skepticism Hacker News commenters scrutinized the GLM 5.2 benchmark, arguing it pits a bare prompt against Semgrep's scaffolded multi-agent pipeline. discussion