Apple Cracks Down on AI Agent Data Access — Builders Must Adapt
Apple restricts full-disk access for AI agents, forcing developers to rethink data pipelines. Plus: new tools, research, and industry takes.
The story of the day
Apple’s latest macOS security update quietly neutered full-disk access for AI agents, a move that directly targets the growing class of autonomous tools scraping local files for context. According to Ars Technica, the change requires explicit user approval for each app—no more blanket permissions. For builders shipping agentic workflows, this means a new friction point: either redesign for granular data access or watch adoption stall. The irony? Apple’s own on-device AI pushes rely on the same data pipelines. Expect a wave of patch notes and user complaints.
New tools & launches
Rowan — Open-source SAST scanner for AI apps (code, models, MCP). [GitHub]
Hive — Simplified, cheaper AI agent API for small teams. [HN]
Models & research
Anthropic committed $100M to its Claude Frontier Academy, aiming to train 10,000 "Frontier Development Engineers" by 2028. The proprietary-focused program signals a talent land grab: if closed models dominate, the skill premium shifts from prompt engineering to system integration. [Unite.AI]
Industry pulse
Chamath Palihapitiya argues open models are a "commoditize your complement" play—give away the model, monetize the adjacent (ads, cloud). The theory holds for Meta/Google but falters for pure-play AI labs: users keep paying for closed models anyway. [Social Capital Research]
Stat of the day
Without search, even top models get 25% of answers wrong. [Randal Olson]
Get the full analysis — subscribe to Gradient Pro.
