GitHub Security Lab's AI Framework Found 80+ Vulnerabilities
GitHub Security Lab's open-source AI framework found 80+ real vulnerabilities by teaching LLMs to understand threat models first. Here's how it works and what it found.
ai-coding
GitHub Copilot Code Review: 60M Reviews and the Agentic Shift
GitHub's Copilot code review hit 60M reviews by rebuilding with an agentic architecture. It now handles 1 in 5 GitHub code reviews, stays silent 29% of the time, and helped WEX ship 30% more code. Here's how the new system works.
GitHub and Andela: Scaling AI Access for Global Developers
GitHub and Andela trained 3,000 engineers on Copilot by embedding AI learning directly into production workflows. The result: 50% productivity gains and proof that the AI skills gap is about access, not ability.
GitHub Copilot Dev Days: Hands-On AI Coding Events Go Global
GitHub is launching free, hands-on Copilot training events globally starting March 2025. Learn AI-assisted coding from GitHub Stars, MVPs, and community experts through practical workshops covering CLI, Cloud Agent, and real workflows.
GitHub Copilot CLI: From Intent to Pull Request in Your Terminal
GitHub Copilot CLI turns natural language into reviewable diffs without leaving your terminal. Use /plan to explore, approve concrete changes, and move from intent to pull request faster.
GitHub Copilot Coding Agent: Five Updates That Actually Matter
GitHub Copilot coding agent now picks models per task, self-reviews its code, runs security scans during work, supports custom agents, and hands off between cloud and CLI without losing context. Here's what changed and why it matters.
How Cline Climbed from 47% to 57% on Terminal Bench in One Weekend
Cline jumped from 47% to 57% on Terminal Bench in one weekend using hill climbing: run evals, diagnose failures, fix one thing, measure again. Here's the exact process, including the Harbor + Modal setup that makes it practical.