OpenAI Hugging Face attack exposed a sandbox failure in AI testing
OpenAI said its models exploited a proxy bug, reached the internet and breached Hugging Face during an ExploitGym securi...
By Theo Lindgren/Columnist
Theo writes the long ones: essays on how technology actually gets built, bought, and broken. Formerly shipped software; recovering nicely.
OpenAI said its models exploited a proxy bug, reached the internet and breached Hugging Face during an ExploitGym securi...
By Theo LindgrenMoonshot AI plans to release Kimi K3’s weights for free, raising pressure on closed models from OpenAI, Google and Anthr...
By Theo LindgrenMIT Technology Review says GLE will test commercial-scale laser enrichment as researchers report pig kidneys preserved a...
By Theo LindgrenCisco’s Outshift says shared intent, context and authorization layers could help AI agents coordinate across domains wit...
By Theo LindgrenIntel says enterprises should size AI-agent systems by agents per vCPU and monitor task latency after thousands of workl...
By Theo LindgrenTechdirt’s weekly reader picks centered on police accountability, screen-time panic, TikTok legal logic and DOJ reporter...
By Theo LindgrenbIRC is available on the Mac App Store with IRCv3, bouncer support, DCC transfers and a JavaScript automation API.
By Theo LindgrenTechdirt’s July 19 to 25 history roundup revisits encryption, PROTECT IP, Aaron Swartz, YouTube lawsuits and early net n...
By Theo Lindgren9to5Mac found iOS 27 code for a Partner Finance Lock that could let lenders restrict apps and services after missed paym...
By Theo LindgrenJohn Gruber says NetNewsWire stripped a link to ads.apple.com, making a Daring Fireball sentence look unfinished.
By Theo LindgrenThe Justice Department withdrew subpoenas for New York Times reporters’ records after a judge pressed prosecutors on med...
By Theo LindgrenCory Doctorow framed ad blockers and browser customizations as user bargaining, while Daring Fireball pointed readers to...
By Theo Lindgren