2026-09-14 · MondayPublished daily at 08:00
AI Daily
- 1An Anthropic report says the Houthi organization used Claude Code to develop missile guidance softwareHacker News: Popular AI posts
- 2
- 3Gary Marcus Reviews Dario Amodei's Proposal to Slow AI Development: Three Points of Praise and Two of SkepticismGary Marcus:The Road to AI We Can Trust
01Industry News
An Anthropic report says the Houthi organization used Claude Code to develop missile guidance software
Anthropic's September threat report disclosed that, according to its assessment, a Yemeni group very likely linked to the Houthi organization used Claude Code to develop software related to guidance rockets, ballistic missiles with ranges exceeding 2,000 kilometers, and a hypersonic glide vehicle concept named R2000.
Original text02Tips and perspectives
An analysis of context engineering for long-running Agent tasks: using budget control, compression, todo-state, and memory to combat context overflow and goal loss
The article analyzes four mechanisms the Agent harness layer uses to address context overflow and goal loss in long-running tasks: context budgeting and offloading, compression, todo-state recitation, and cross-session memory.
Original textGary Marcus Reviews Dario Amodei's Proposal to Slow AI Development: Three Points of Praise and Two of Skepticism
Gary Marcus critiques Dario Amodei's article advocating a slowdown in AI development and greater transparency, which Sam Altman and Elon Musk quickly endorsed. Marcus praises the transparency commitment but lists several concerns: METR's close ties to AI companies, Anthropic invoking the threat from China to sustain acceleration, and the possibility that the proposal is intended to preempt genuine regulation.
Original text