Claude Code Wipes C Drive: Why Skipping Permissions Led to Disaster
Developer Zac lost his entire C drive after running Claude Code Opus 5.5 with --dangerously-skip-permissions, but daily NAS backups restored 98% of data; the incident sparked debate over Anthropic's Auto Mode safety classifier, which blocks 89% of dangerous commands, yet a Linux user still lost files even with Auto Mode enabled.
Developer Zac shared on X that Anthropic's Claude Code Opus 5.5 deleted his entire C drive while running autonomously for hours. The post garnered over 1 million views. Zac had used the --dangerously-skip-permissions flag to prevent the agent from pausing for authorization, a workflow he had used since Opus 4.6 without issues. He suspects a PowerShell syntax concatenation error caused the destructive command.
Backup Saved 98% of Data
Zac credited daily automatic backups to a Synology NAS for recovering approximately 98% of his data. He acknowledged personal responsibility, stating, "This is nobody's fault but my own. We're all still in the 'Wild West' of this evolving AI world." However, he argued that baseline protection against whole-disk deletion should not rely solely on user-configured backups.
Anthropic Recommends Auto Mode
Claude Code lead Boris Cherny responded on X, noting this is exactly why Anthropic recommends Auto Mode (default for Pro, Max, and Team users since August 14, 2026). Auto Mode routes tool calls through a dedicated classifier that intercepts dangerous, irreversible, or out-of-scope actions. If blocked, Claude attempts a safer approach or asks for approval. Internal testing with 1,053 paid testers showed humans caught only 13.6% of dangerous commands, while Auto Mode intercepted 89%.
Why Zac Avoided Auto Mode
Zac explained that Auto Mode frequently interrupts tasks, forcing repeated manual intervention — "like babysitting" — preventing true hands-off operation. Anthropic disclosed that about 25% of interactive Claude Code sessions start with permissions fully skipped, indicating many users prioritize efficiency over safety guards.
Community Debate: Windows, PowerShell, or User Error?
Some blame Windows and PowerShell's complex piping and path handling, noting LLMs can mishandle slashes in generated commands.
Others question Zac's project isolation: whether he used a dedicated working directory and properly scoped the agent.
A few suggest the issue reflects Zac's operational practices rather than the model.
Linux User Loses Data Even With Auto Mode
A Linux user reported a similar data-loss incident last week while using Auto Mode. The cause was not a malformed command but a flawed test written by Claude Code for filesystem code in the project. The user recovered via full backups and dotfiles-managed configs, but emphasized that Auto Mode is not foolproof and the problem is not Windows-specific.
Anthropic's Own Documentation Warns
Official docs state Auto Mode reduces but does not eliminate risk, and recommend running agents in isolated environments whenever possible.
Contrast: Successful C Drive Cleanup
One commenter shared a positive experience: asking an AI to clean the C drive freed 65 GB of orphaned caches and forgotten files. Yet the article concludes that the core challenge is no longer whether AI can write code, but whether errors can be intercepted in time and data recovered — especially since not every developer has a NAS backing them up.
References: https://madrobot.blog/2026/10/07/claude-opus-5-5-deleted-c-drive-claude-code-auto-mode-boris-cherny/ , https://x.com/PerceptualPeak/status/2107621483392446572 .
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
IT Services Circle
Delivering cutting-edge internet insights and practical learning resources. We're a passionate and principled IT media platform.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
