Hacks that exploit flaws in crypto software are running at 2.5 times their 2022–2025 pace, starting five weeks after OpenAI and Anthropic released new coding agents. Code flaws now account for nearly two-thirds of reported losses.
Research ·
In September, OpenAI’s GPT-6 Astra drew OpenAI level with Anthropic in AI model spending. Three weeks after launch, Claude Opus 5.5 has pushed Anthropic back to its pre-Astra share on all three public trackers we monitor.
Research ·
Published software vulnerabilities rose 106% from January through September compared with a year earlier. Those with evidence of real attacks rose 34%.
Research ·
New math papers on arXiv, the main open archive for math, physics and computer science research, are up 78% from a year ago. Biology papers show no such acceleration.
Research ·
Shopify product catalogs turned up behind sites from all five AI builders we studied. On Lovable, about one in six stores pulls its products from Shopify.
Research ·
OpenAI passed Anthropic in spending on OpenRouter and in Ramp’s billing sample in September, and is closing in on Vercel’s AI Gateway. Model-level data trace the gain to GPT-6 Astra.
Research ·
Village’s study of 25.7 million web hosts reveals how three shared platforms can reverse Datadog’s apparent lead over New Relic.
Research ·
Across a crawl of 20.7 million hosts, Village identified 87,003 live sites managed by Lovable, Replit, Base44, v0, and Bolt. Most of the visible ones are real businesses or working products, yet none rank among the web's 10,000 most-visited sites.
Research ·