Live Feeds
● TRACKER Updated 24d ago · 39 sources tracked

48,000 files deleted in 103 seconds: Claude Code just destroyed a developer’s work at the click of a button, and it shows why everyone should be careful around AI agents

Security researchers uncovered multiple vulnerabilities and threat campaigns targeting the Claude Code AI tool, exposing developers to severe risks. Noma Labs discovered a desktop application flaw allowing malicious HTML files to steal local credentials and private keys through the preview pane. Security firm Noma Security previously identified an 8.7-severity vulnerability named SneakPeek that bypassed sandboxing. Additional threats include a September 2026 incident where a Windows junction caused the tool to delete 48,218 project files in 103 seconds, alongside separate malware payloads running the software with permissions disabled.

🎙️

Listen to Live Briefing

Real-time synthesized voice briefing · Live Feeds Desk

⏱ ~3 min
Speed:
RSS Source map (49)
⚡ Key Developments & Real-Time Context
Text size:
  • ✓ Noma Labs found a high-severity Claude Code Desktop flaw that let malicious HTML files steal local credentials and private keys through the preview pane.
  • ✓ A Windows junction caused Claude Code to delete 48,218 project files in 103 seconds in September 2026.
  • ✓ Noma Security identified a high-severity flaw in the desktop application rated 8.7 out of 10 called SneakPeek.
🛡️ Source Corroboration: 39 independent reporting domains (95% confidence) ⏱ Read time: ~2 min

What changed

Noma Labs identified a desktop flaw allowing malicious HTML preview files to harvest local credentials and private keys.

Live updates

  1. Claude Code Flaws Expose Credentials and Trigger Mass Deletions

    Security researchers uncovered multiple vulnerabilities and threat campaigns targeting the Claude Code AI tool, exposing developers to severe risks. Noma Labs discovered a desktop application flaw allowing malicious HTML files to steal local credentials and private keys through the preview pane. Security firm Noma Security previously identified an 8.7-severity vulnerability named SneakPeek that bypassed sandboxing. Additional threats include a September 2026 incident where a Windows junction caused the tool to delete 48,218 project files in 103 seconds, alongside separate malware payloads running the software with permissions disabled.

    Why it matters

    AI coding assistants are increasingly integrated into developer workflows, granting deep system access that makes them high-value targets for attackers. Recent campaigns utilize fake advertisements and ClickFix social engineering to compromise enterprise employees and individual developers. These incidents highlight the security dangers of autonomous development tools operating with elevated permissions.

    What is confirmed

    • Noma Labs found a high-severity Claude Code Desktop flaw that let malicious HTML files steal local credentials and private keys through the preview pane.
    • A Windows junction caused Claude Code to delete 48,218 project files in 103 seconds in September 2026.
    • Noma Security identified a high-severity flaw in the desktop application rated 8.7 out of 10 called SneakPeek.

    Still unconfirmed

    • Threat actors are impersonating Claude and ChatGPT with ClickFix-style social engineering to compromise developers and enterprise employees.
    • In 2025 the s1ngularity payload launched Claude Code, Gemini CLI and Amazon Q with approval prompts off.

    What to watch next

    • Releases of official patches addressing the Noma Labs desktop preview vulnerability
    • Security audits focusing on sandbox isolation for AI development tool preview panes
    Sources used for this update (5)
    1. hackread.com — Claude Code Desktop Flaw Let Malicious HTML Steal Developer ...
    2. dev.to — The Nx Malware Ran Claude Code With --dangerously-skip ...
    3. www.lowcode.agency — Claude Code Source Code Leaked? Here’s what it contains
    4. securityarsenal.com — Fake Claude and ChatGPT ClickFix Campaigns Targeting ...
    5. www.dualmedia.com — Fake Claude Ads Exploit Google and Bing Redirects
    confidence 95%

  2. CyberXero uses Claude Code in Ukraine data theft as security flaws emerge

    The AI coding tool Claude Code is linked to a campaign by CyberXero that exposed over 628,000 Ukrainian records. Separately, security firm Noma Security identified a high-severity flaw in the desktop application rated 8.7 out of 10. This vulnerability, termed SneakPeek, allows poisoned HTML files to steal AWS, GitHub, and SSH credentials without user interaction by bypassing sandboxing layers. These developments follow a September 2026 incident where a Windows junction caused the tool to delete 48,218 project files in 103 seconds.

    Why it matters

    These incidents highlight risks associated with autonomous AI agents having direct system access. The transition from accidental data loss to active exploitation by initial access brokers indicates a growing threat vector for developers.

    What is confirmed

    • CyberXero used Claude Code in a campaign that stole over 628,000 Ukrainian records.
    • A vulnerability called SneakPeek allows poisoned HTML files to steal AWS, GitHub, and SSH credentials without a click.
    • Noma Security rated a Claude Code desktop application flaw 8.7 out of 10 for severity.
    • A Windows junction led a Claude Code sub-agent to delete 48,218 files in 103 seconds.

    Still unconfirmed

    • Fake AI tools targeting US companies are impersonating Claude, ChatGPT, and Copilot.

    What to watch next

    • Anthropic's official response or patch notes regarding the SneakPeek vulnerability
    • Further attribution of the CyberXero campaign by international security agencies
    Sources used for this update (7)
    1. www.esecurityplanet.com — CyberXero Uses Claude Code in Ukraine Data Theft
    2. aigovernance.com — Claude Code Flaw Let Malicious Files Steal Developer…
    3. www.cyberincidents.in — CyberXero Uses Claude Code in Campaign That Stole Over ...
    4. dev.to — 48,218 Files in 103 Seconds: How a Windows Junction Let an ...
    5. dev.to — The Claude Code Desktop Vulnerability That Beat Three Layers ...
    6. www.penligent.ai — Claude Code SneakPeek Vulnerability: How a 256KB HTML Preview ...
    7. any.run — Fake AI Tools Target US Companies: Claude, ChatGPT & Copilot
    confidence 90%

  3. AI Coding Tool Claude Code Deletes 48,000 Files in 103 Seconds

    An autonomous AI coding tool, Claude Code, deleted 48,218 project files and associated Git history in 103 seconds without user approval prompts. The incident occurred in September 2026 while a developer was building a testing environment for stock options analysis software. The tool mistakenly targeted production files instead of test copies, causing irreversible data loss due to the lack of separate backups.

    Why it matters

    This incident highlights the risks associated with giving AI agents system access and the importance of robust safeguards, such as sandboxing and regular backups. The developer's loss underscores the need for careful oversight and control when using autonomous AI tools. The incident has sparked discussions on AI safety and the measures needed to prevent similar disasters.

    What is confirmed

    • Claude Code deleted 48,218 project files and associated Git history in 103 seconds.
    • The incident occurred in September 2026.
    • The developer had no separate backup of the project files.
    • CyberXero used Claude Code in a campaign that stole over 628,000 Ukrainian records.

    Still unconfirmed

    • Claude Code can delete files without approval prompts if not properly configured.

    What to watch next

    • Further investigation into Claude Code's safety features and user controls
    • Development of more robust safeguards for AI coding tools
    • Outcome of the parliamentary audit on South Korean bank breaches linked to exposed Claude Code sessions
    Sources used for this update (17)
    1. www.xvideos.com — blowjob videos - XVIDEOS.COM
    2. blowjobs.pro — Best Blowjob Videos, Girls Sucking Dick | Blowjobs.pro
    3. www.xvideos.com — best-blowjob videos - XVIDEOS.COM
    4. www.xnxx.com — Blowjob videos - XNXX.COM
    5. www.pornhub.com — Blowjob Porn Videos | Pornhub.com
    6. blowjobit.com — Blowjob Video Categories - The best free blowjob videos |...
    7. blowjobs.pro — Free Blowjob Videos by Categories | Blowjobs.pro
    8. tubepleasure.com — Blowjob - Tube Pleasure
    9. phronews.com — How Claude Code Allegedly Deleted 48,000 Files in Seconds
    10. dev.to — Claude Code can delete your files. The 3-layer setup I run ...
    11. ethics.ai — AI incident timeline | ethics.ai
    12. ccleaks.com — ccleaks - Claude Code & AI supply chain news
    confidence 90%

  4. Anthropic AI Agent Wipes 48,000 Files in Seconds

    An autonomous AI coding tool wiped 48,218 project files and the associated Git history in just 103 seconds during a September 2026 incident. The software executed the deletions without requiring approval prompts from the user. The disaster occurred while the developer attempted to build a testing environment for stock options analysis software. The tool mistook Windows junction folders for normal directories, following those system pointers straight into production files instead of localized test copies. The developer possessed no separate backup.

    Why it matters

    Autonomous coding assistants frequently execute system commands without human oversight, creating high-risk scenarios when permission guardrails are missing. This incident demonstrates how automated tools can misinterpret file structures and cascade destructive actions across local storage. Security specialists are urging developers to alter configuration settings and implement sandboxing before deploying agents to modify codebases.

    What is confirmed

    • Anthropic's Claude Code deleted 48,218 project files in 103 seconds during a September 2026 incident.
    • The AI agent ran without approval prompts.
    • The incident also resulted in the deletion of the project's Git history.

    Still unconfirmed

    • Four specific configuration settings should be altered before allowing an agent to touch code.

    What to watch next

    • Adoption of mandatory sandboxing and backup protocols for AI coding tools
    • Developer community responses regarding default safety guardrails in AI agents
    Sources used for this update (3)
    1. bluemediaworld.com — Claude Code Deleted 48,000 Files in 103 Seconds — Here’s How ...
    2. dueclix.com — AI Agent Deleted 48,000 Files: What a Small Team Should ...
    3. airviorg.substack.com — 48,000 files, 103 seconds, no guardrails - AIRVI
    confidence 95%

  5. Claude Code deleted 48,000 files due to Windows junction folder error

    Anthropic's Claude Code AI agent deleted 48,218 files in 103 seconds while attempting to build a testing environment for stock options analysis software. The agent successfully completed 10 of 11 repair tasks before failing on a final request to create a mirror environment. The AI misidentified 614 Windows junction folders as standard directories and followed those pointers to delete production files instead of the test copies. The developer had no backup of the destroyed project.

    Why it matters

    This incident demonstrates a failure in how AI agents interact with system-level file structures. It highlights the danger of granting AI tools broad write permissions without strict environment isolation.

    What is confirmed

    • Claude Code deleted 48,000 files in 103 seconds.
    • The AI agent was working on stock options analysis software.

    Still unconfirmed

    • The testing environment contained 614 junction folders pointing to live working files.

    What to watch next

    • Anthropic's official response or patch regarding Windows junction folder handling.
    • Developer community reports of similar file system errors with Claude Code.
    Sources used for this update (2)
    1. aipowerranking.com — Claude Code AI Agent Deletes 48,000 Files in 103 Seconds ...
    2. specpilot.dev — Claude Code Deleted 48,000 Files in 103 Seconds. The Missing ...
    confidence 90%

  6. Claude Code AI Agent Deletes 48,000 Files in 103 Seconds

    A developer's project was destroyed when an AI agent using Claude Code deleted approximately 48,000 files in 103 seconds. The incident occurred when the developer instructed the AI to modify a test copy, but it mistakenly identified and deleted files from the actual working directory. The AI's actions were not restricted by safety instructions, which were merely prompts, not enforced system-level permissions.

    Why it matters

    This incident highlights the risks associated with allowing AI agents to execute commands with broad permissions. As AI agents become increasingly capable of writing code, running commands, and managing complex development tasks, the potential for disasters like this one grows. The incident serves as a warning about the importance of carefully configuring and monitoring AI agents.

    What is confirmed

    • Claude Code deleted approximately 48,000 files in 103 seconds.
    • The AI agent was instructed to modify only a test copy, but it deleted files from the actual working directory.
    • The incident was caused by the AI following Directory Junctions, which are Windows mechanisms that act as 'portals' linking to directories elsewhere.
    • The AI also cleared the local `.git` objects, refs, and logs directories, destroying the version history.

    Still unconfirmed

    • The developer had the AI agent run 11 tasks, and one of them caused the deletion.

    What to watch next

    • Further investigation into the safety features of Claude Code and other AI agents
    • Development of more robust safety instructions and system-level permissions for AI agents
    • Decisions from Anthropic, the creator of Claude Code, on how to prevent similar incidents in the future
    Sources used for this update (10)
    1. Yahoo Tech — 48,000 files deleted in 103 seconds: Claude Code just destroyed a developer’s work at the click of a button, and it shows why everyone should be careful around AI agents
    2. CyberSecurityNews — Claude Code Agent Allegedly Deletes 48,000 Files in 103 Seconds
    3. Android Headlines — A Claude Code Rampage Allegedly Wiped Out 48,000 Project Files in Under Two Minutes
    4. techradar.com — ‘Do not let programs run commands on your machine’: Claude Code allegedly deleted 48,000 files in 103 seconds, and it’s a terrifying warning about AI agents
    5. finance.biggo.com — Claude Deletes 48,000 Real Files in 103 Seconds—AI Agent Safety Boundaries Under Fresh Scrutiny
    6. currently.att.yahoo.com — 48,000 files deleted in 103 seconds: Claude Code just ...
    7. www.htx.com — 103 Seconds, Claude Deleted 48,000 Real Files. 'Don't Touch ...
    8. www.gadgetgeeksters.com — Claude Code AI Agent Deleted 48,000 Files in 103 Seconds ...
    9. currently.att.yahoo.com — 'I broke something': A Claude Code AI agent deleted 48,000 ...
    10. sigmawire.net — Claude Code Deleted 48,000 Files: What Happened
    confidence 95%
📊

Community Sentiment: How do you assess this situation?

Voice your perspective · Real-time aggregated sentiment from the Live Feeds community