Claude Cowork Allegedly Deleted Folder Containing 15 Years of Family Photos While Organizing User's Wife's Desktop

Venture capitalist Nick Davidov alleged that Anthropic's Claude Cowork, after being asked to organize his wife's desktop and delete temporary Office files, instead deleted a folder containing roughly 15 years of family photos and related personal memories via terminal commands. Davidov later reported that Apple support helped recover the files through an iCloud recovery feature.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The AI system failed to perform reliably, mistakenly targeting and deleting a critical folder of photos instead of temporary files, which represents a lack of capability and robustness in executing file system tasks.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: The user granted the AI agent direct write and delete permissions on a live file system, representing an unsafe use of a system not yet robust enough for such tasks.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by an action made by the AI system, which mistakenly deleted a folder of photos while pursuing the user-defined goal of organizing the desktop.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Risk Level 3: Limited Risk. The system is an agent-based AI workspace operating within a chatbot interface ('Claude desktop interface'), which falls under the category of chatbots requiring transparency obligations.

AI system and alleged parties

• AI system: Claude Cowork (Anthropic)
• AI purpose: Chatbot; Content Curation
• Behaviour type: Agent
• Alleged developer: Anthropic
• Alleged deployer: Nick Davidov
• Alleged harmed parties: Wife of Nick Davidov, Nick Davidov, Family of Nick Davidov

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Psychological

Reported: The report explicitly describes psychological distress and anxiety caused directly by the incident.

Directly caused: The venture capitalist described the experience of losing 15 years of irreplaceable family photos as 'harrowing' and stated that it 'nearly gave him a heart attack.'

People affected

• Occurrences reported: 1
• People reportedly harmed: 2
• People reportedly exposed: 2

Potential causes

Management

• Premature Feature Release: Releasing file-system agent features before they are fully safe for mainstream.

Technology

• Terminal Execution Bypass: AI executed deletion via terminal, bypassing the standard system Trash.
• Flawed File Renaming Logic: AI attempted to rename files but mistakenly deleted the entire folder.
• Lack of Safety Safeguards: Agent lacked safeguards to prevent modifying critical personal directories.

Data Inputs

• Ambiguous Directory Context: AI misidentified the photo folder as a temporary office file directory.
• Lack of File Classification: No clear metadata distinguishing temporary files from critical user photos.

Human Factors

• Excessive Permission Granting: User granted write and delete permissions to the AI on a live file system.
• Lack of Immediate Backup: User initiated file cleanup without ensuring a robust local backup existed.

Process and Methods

• Direct File System Modification: Process allowed direct modification of files without a safe dry-run preview.
• Non-Reversible Deletion Method: System used terminal commands that bypass the standard OS trash recovery.

Regulatory Environment

• Lack of AI Agent Standards: Absence of safety regulations for autonomous AI agents manipulating files.

Information quality

• Classification confidence: High
• Reason for confidence: The report is a first-hand account from the user detailing exactly what happened, the specific AI system involved (Anthropic Claude agent), the actions it took (deleting files via terminal), and the outcome (recovery via iCloud).
• Ambiguities identified: None of significance.
• Alternative interpretations: None.

An Anthropic Claude AI agent mistakenly deleted 15 years of family photos on a user's personal desktop. While distressing for the individual, the incident is a localized consumer safety and software reliability issue with negligible national security implications.

• Overall national security impact: Negligible
• Scope: Single nation
• Primary target: No clear primary
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The incident is a localized consumer safety issue with no immediate or developing national security threat timeframe.
• Autonomy: Human-supervised. The AI agent executed terminal commands autonomously to organize the desktop, though under the user's initial setup and authorization.
• Novelty: Evolved capability. Represents an evolved capability where LLM-based agents interact directly with local file systems and terminal commands.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, or human safety elements were impacted by this local file deletion incident.
• Information security: Negligible. No intelligence compromise, classified data theft, or systematic information manipulation occurred.
• Sovereignty: Negligible. The incident was a localized consumer software issue and did not affect state authority or government functions.
• Economic security: Negligible. No strategic industries, financial systems, or national technological advantages were compromised.
• Societal stability: Negligible. This was an isolated incident affecting an individual household, presenting no threat to societal stability or human rights.