AudioEye tested leading AI coding tools and found they consistently produce inaccessible websites. Meanwhile 81% of dev teams believe they're compliant.

RuleEvolve is a self-evolving framework that iteratively improves AI coding agent rules, outperforming hand-crafted prompts on correctness, code length, and token cost.

YTL AI Labs launched ILMUCode, an agentic AI coding platform built on its own ILMU-GLM-5.3 language model and sovereign infrastructure, developed with Z.ai.

Artificial Analysis benchmarks pit Claude Code with Sonnet 5.5 against Codex with GPT-6.1 Sol. The scores are close. The cost gap is not.

Querit's Search API now supports a dedicated code vertical with a single parameter change. It scored 83.17% on FreshQA and returned directly usable results in 81% of coding tasks.

AIR Security found that Claude Code, GitHub Copilot, Gemini, and OpenAI Codex skip post-checkout hash verification, letting attackers swap in malicious code undetected.

NVIDIA launched OpenShell on September 28, putting AI coding agents inside policy-controlled sandboxes that block credential exfiltration at the system level, not the prompt level.

Glow Labs found 13,000 internal company images in 900+ public GitHub repos across 300+ organizations. Billing records, unreleased features, and treasury consoles were exposed.

Ahrefs outlines the core reasons brands lose visibility in AI-powered search results, from missing conversations to content AI assistants can't access.

Ahrefs mapped out 35 ways to run SEO work inside Claude Code, from content briefs and Search Console analysis to internal linking and technical fixes.

A controlled study of 1,116 AI-built web apps found a boot probe removes nearly all launch failures at 35% of a full shell's token cost. Screenshots help less than expected.

JetBrains surveyed 15,000+ developers and found AI agents now write nearly half of all professional code. Here are the numbers on tool share, trust, security, and revenue.