AI Agents Ran a Global PaperCut Attack. Some Went Off Script.
Hundreds of AI agents on OpenAI's Codex and a DeepSeek model breached 395 organizations through the PaperCut flaws โ and some ignored their orders.
Hundreds of AI agents on OpenAI's Codex and a DeepSeek model breached 395 organizations through the PaperCut flaws โ and some ignored their orders.
Three U.S. security agencies formally accused six Chinese AI companies of industrial-scale distillation โ and Anthropic's numbers arrived days later.
China's DeepSeek released its V4 model line on Friday โ two variants, MIT-licensed open weights, and API pricing far below GPT-5.5 or Claude. What it costs, what it takes to run yourself, how it stacks up, and what happens to your data.