AINewsnow

91% Passed, 25% Safe: Why Your Coding Agent Ships Code That Works and Gets You Hacked

A new benchmark took 150 real vulnerabilities, handed the broken code to six frontier agents, and graded the output the way an attacker would. The best agent got 91.1% of the tasks working. Only 25.1% of them were also safe to ship. The Java config importer that passed every test One task in the be…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-10-16 16:03 · DEV Community — Machine Learning
    91% Passed, 25% Safe: Why Your Coding Agent Ships Code That Works and Gets You Hacked

More stories

  1. NVIDIA RTX Spark Laptops Launch Globally with AI Agent Support — NVIDIA Blog
  2. Reflection AI releases open-weight 501B model Beam for agentic coding — AWS Machine Learning Blog
  3. Meta and Apple collaborate to enable Muse Spark Open AI model on Macs — Wall Street Journal Technology
  4. Google Gemini 4 Argon Demonstrates Advanced Physical Reasoning and Robotics — Google DeepMind YouTube
  5. Epic Games releases new launcher built with AI assistance — Ars Technica AI
  6. New Documentary Examines Elizabeth Holmes and Her Impact on Tech — Hard Fork (NYT)
  7. Zephon brings elastically deterministic data loading to TorchTitan — PyTorch Blog
  8. Meta releases Muse Spark Open 816B model; NVIDIA Nemotron 3.5 gains traction — AWS Machine Learning Blog

Get the daily brief of stories like this at 6:30 every morning →