AI News
ScopeBench Tests Whether AI Hacking Agents Know When to Stop
September 29, 2026
A new benchmark called ScopeBench builds 30 security tasks with no legal way to win: the flag only exists behind a boundary agents were told not to cross. Eight AI models were tested across 2,160 trajectories, and the gap between hacking skill and actually respecting scope turned out to be a lot wider, and more troubling, than anyone might have guessed.
Read more
NaiveAI Releases Naive-N0.5-Flash, an Open-Weight 309B MoE Model With Native 1M-Token Context
September 28, 2026
NaiveAI has published final weights and a model card for Naive-N0.5-Flash, a 309-billion-parameter open-weight model with just 15.5 billion active parameters, a native 1M-token context window, and API pricing starting at a cent per million cached tokens.
Read more
ParseBench Wants to Know If Your PDF Parser Is Lying to Your AI Agent
September 27, 2026
ParseBench is a new open-source benchmark testing whether document parsers preserve the structure AI agents need, not just whether the output looks visually correct. LlamaParse Agentic Plus currently tops its 2,078-page leaderboard at 90.20.
Read moreAnalysis
Guides
Cross-Post 5 Platforms in 15 Minutes: The Typefully Workflow
May 22, 2026
Native posting across 5 platforms = 60 minutes. Typefully cross-posting = 15. The 6-step workflow that saves 20 hours/month, and the 4 common mistakes to skip.
