Claude Code Scans Local Drives—Anthropic Admits Data Overreach
Anthropic confirms Claude Code scans entire drives—builders question data boundaries. Plus: new tools for AI workflows.
The story of the day
Anthropic confirmed via GitHub issues that Claude Code scans entire local drives by default, only disclosing the behavior when explicitly prompted. For builders deploying AI agents in sensitive environments, this raises a familiar tension: convenience vs. control. The admission follows a pattern of post-deployment transparency, where edge cases in tool behavior surface only after user friction. The real question isn’t whether the scan is intentional—it’s whether the default should be opt-in, not opt-out.
New tools & launches
StepYard — Open-source local automation runner with YAML pipelines for LLM-driven workflows. [GitHub]
Logslim — Open-source tool to compress test/build output before feeding it to AI agents. [GitHub]
StayUp — Keeps Macs awake (lid closed) during long AI agent runs. Free. [Website]
Blogr.ai — AI-generated SEO action plans from Search Console data. [Website]
Models & research
No new model benchmarks or papers surfaced in the last 24h. The quiet suggests a lull—or a pivot to closed-door deployments.
Industry pulse
O’Reilly’s Radar questions code ownership for AI-generated output, highlighting the gap between Anthropic’s ToS and real-world developer expectations. The debate mirrors open-source licensing tensions: who controls the IP when the "author" is a model fine-tuned on proprietary data?
Stat of the day
Claude Code’s GitHub issue thread on local scans has 3x more reactions than any other recent Anthropic repo discussion. [Source]
Get the full analysis — subscribe to Gradient Pro.
