48/100 CVSS 7.5 🔴 CVE ANALYSISPublished: July 24, 2026

Claude Code: Sandbox Escape via Git Worktree Path Confusion Allows Unsandboxed Code Execution

Claude Code's worktree handling allowed creation of worktrees named ".git" and navigation to worktrees outside the sandbox context, enabling git directory confusion attacks. By exploiting symlink manipulation and git fsmonitor execution during worktree operations, an attacker could overwrite files in the user's home directory (such as .zshenv), lea...

7.5

CVSS Score

MEDIUM

Threat Level

Priority /100

TBD

Exploited ITW

Monitor

CISA Status

Sources

🔵

HIGH-PRIORITY SECURITY ADVISORY — Priority Score: 48/100

CVSS 7.5 MEDIUM. SENTINEL APEX recommends immediate patch evaluation. Intelligence from 1 confirmed source(s).

📋 Executive Summary

⚡ Analyst Assessment — SENTINEL APEX v4.0

This report analyzes CVE-2026-55607 affecting npm @anthropic-ai/claude-code (CVSS 7.5, MEDIUM severity). No confirmed in-the-wild exploitation at the time of writing; prioritize on exposure, privilege, and CVSS. Corroborated across 1 source(s): github_advisories. Verify all specifics against the primary sources linked below before acting.

Intelligence sources: github_advisories

🧬 Weakness Anatomy — CWE-22, the underlying flaw class

This vulnerability is classified as CWE-22: Path Traversal in the primary record. The mechanics below are characteristic of this weakness class (per the MITRE CWE corpus), not a claim about a specific exploit observed against your environment.

How it's exploited	Attacker uses ../ sequences or absolute paths to access files outside the intended directory, reading or writing arbitrary files.
Detect & prevent	Canonicalize and confine paths to a safe root; reject traversal sequences. Detect via web logs for ../ and encoded-traversal patterns.

⚠️ Business Impact Analysis

⚠️ npm @anthropic-ai/claude-code users face material security risk requiring immediate remediation

🔗 Representative Attack Path

A typical attack path for this vulnerability class, mapped to MITRE ATT&CK. This is a representative model to guide detection and defense — not a claim of observed activity against a specific target. Confirm specifics against the primary sources below.

#	Phase	Attacker Action	MITRE
1	Reconnaissance	Attacker identifies exposed @anthropic-ai/claude-code instances via Shodan, Censys, or targeted scanning	`TA0043`
2	Initial Access	Exploitation of CVE-2026-55607 in npm @anthropic-ai/claude-code	`T1190`
3	Execution	Remote code execution achieved — attacker runs arbitrary commands on target system	`T1203`
4	Persistence	Backdoor, scheduled task, or new admin account created for persistent access	`T1053`
5	Command & Control	Attacker establishes persistent C2 channel using HTTPS or DNS tunneling	`T1071`
6	Impact / Objectives	Intellectual property theft, espionage, cryptomining, or preparation for future attack stage	`T1657`

⚠ Deep Dive Analysis

This high-severity vulnerability in npm @anthropic-ai/claude-code (CVSS 7.5) represents a significant attack surface. SENTINEL APEX assesses exploitation to be technically feasible with moderate effort. Organizations running @anthropic-ai/claude-code in internet-facing or privileged positions face immediate risk.

SENTINEL APEX URGENCY: ELEVATED. Score: 48/100 MEDIUM. Patch before exploitation activity begins.

🧠 Structured Intelligence Assessment

SENTINEL APEX analytical tradecraft separates verified fact from labeled assessment and states intelligence gaps explicitly. Confidence reflects evidentiary support, not certainty.

✔ Verified Facts

Affected CVE(s): CVE-2026-55607, CVE-2026-55607.
Affected technology: npm @anthropic-ai/claude-code.
CVSS base score of 7.5 assigned by an authoritative source.
1 indicator(s) of compromise extracted from source reporting.

🕸️ Correlated Observations

Vendor npm has been targeted in 29 prior CYBERDUDEBIVASH reports (recurring target)

🧩 Analyst Assessments

MEDIUM CONF Technical severity is assessed as high (CVSS 7.5); realized business impact depends on asset exposure and data classification.

❓ Intelligence Gaps

In-the-wild exploitation is unconfirmed as of this report.
Attribution to a specific threat actor is not established by the available evidence.
The source lacks sufficient technical detail for confident MITRE ATT&CK mapping.

🔮 Forward Outlook

MEDIUM CONF Given the repeated targeting pattern, further activity against this vendor/sector is probable.

🧬 Prior Intelligence Context

This report is correlated against CYBERDUDEBIVASH SENTINEL APEX's persistent analyst memory. The following entities have been tracked in prior intelligence:

Vendor npm — previously observed 29 times in CYBERDUDEBIVASH intelligence since 2026-07-06 (recent: better-auth-account-takeover-via-pre-account-hijacking-on-m, postcss-path-traversal-in-previous-source-map-auto-loading)

🎯 MITRE ATT&CK Mapping

Category	Mapping
Primary Tactic	ML Model Access
Primary Technique	AML.T0051 — LLM Prompt Injection
Sub-Technique	AML.T0054 — Prompt Injection via Jailbreak
Weakness (CWE)	CWE-22
Intel Type	🔴 CVE ANALYSIS
Source(s)	github_advisories

🔴 CVE Reference

CVE ID	Reference	Score
CVE-2026-55607	NVD →	7.5
CVE-2026-55607	NVD →	7.5

🏷️ Indicators of Compromise — Enriched IOC Feed

Normalized IOCs with confidence scoring. Block immediately across all enforcement points. SOC Pro subscribers receive enriched IOC bundles with full attribution and STIX/TAXII feeds.

Indicator Value	Type	Confidence	First Seen
hackerone.com	domain	75%	2026-07-24

🔍 Detection Rules — Sigma (SIEM)

Deploy to Splunk, Elastic, Microsoft Sentinel, or QRadar. SOC Pro subscribers receive pre-compiled SIEM-native query packs updated with each pipeline run.

Sigma YAMLtitle: CVE-2026-55607 Exploitation Attempt — npm @anthropic-ai/claude-code
id: 03bd94cf879f067f-ba5e
status: experimental
description: Detects exploitation of CVE-2026-55607 in npm @anthropic-ai/claude-code. Priority: MEDIUM. Score: 48/100
author: CYBERDUDEBIVASH SENTINEL APEX v4.0 (bivash@cyberdudebivash.com)
date: 2026-07-24
references:
    - https://nvd.nist.gov/vuln/detail/CVE-2026-55607
    - https://blog.cyberdudebivash.in/
tags:
    - attack.initial_access
    - attack.t1190
    - cve-2026-55607
    - cve-2026-55607
logsource:
    category: webserver
detection:
    keywords:
        - 'CVE-2026-55607'
        - 'CVE-2026-55607'
        - 'cve-2026-55607'
    condition: keywords
falsepositives:
    - Security scanners
level: medium

📡 Detection Rules — YARA (Endpoint)

Deploy to endpoint detection platforms. Enterprise subscribers receive tuned, FP-validated YARA rule packs.

YARArule CVE_2026_55607_Exploitation {
    meta:
        description = "Detects artifacts related to CVE-2026-55607 exploitation in npm @anthropic-ai/claude-code"
        author      = "CYBERDUDEBIVASH SENTINEL APEX v4.0"
        date        = "2026-07-24"
        severity    = "HIGH"
        cvss        = "7.5"
        priority    = "48/100"
        threat_level = "MEDIUM"
        reference   = "https://nvd.nist.gov/vuln/detail/CVE-2026-55607"
    strings:
        $id_str  = "CVE-2026-55607" ascii nocase
        $product = "_anthropic_ai_claude_code" ascii nocase wide
    condition:
        any of them
}

🧠 Multi-Platform Detection Engineering

Evidence-driven detections mapped to MITRE ATT&CK, generated by the SENTINEL APEX Detection Engine — one validated rule per technique across SIEM, EDR, and network platforms. SOC Pro subscribers receive these as deploy-ready rule packs with each pipeline run.

Network Detection — Suricata

Suricata IDS/IPSalert dns $HOME_NET any -> any any (msg:"CYBERDUDEBIVASH SENTINEL APEX - malicious domain hackerone.com"; dns.query; content:"hackerone.com"; nocase; classtype:trojan-activity; sid:9000001; rev:1;)

🎯 Multi-Audience Intelligence Products

This single analysis is packaged for every stakeholder — from the boardroom to the SOC — plus a machine-readable feed for API and MSSP integration. Enterprise & SOC Pro subscribers receive these as structured deliverables.

🏛 Executive Advisory — CISO / Security Leadership

Affected CVE(s): CVE-2026-55607, CVE-2026-55607.
Affected technology: npm @anthropic-ai/claude-code.
CVSS base score of 7.5 assigned by an authoritative source.
1 indicator(s) of compromise extracted from source reporting.

Owner	Decision	Timeline
CISO / SOC Lead	Authorize response activation and detection deployment	Within 24h
Vulnerability Mgmt	Prioritize remediation/patching of npm assets	Within 7 days
Legal / Privacy	Assess regulatory notification obligations if in-scope data is at risk	Within 48h

📋 Board Brief

A high-severity cyber threat affects npm @anthropic-ai/claude-code.
Technical severity is rated 7.5/10.
Primary exposure is operational disruption, data loss, and potential regulatory liability (e.g. GDPR, NIS2, DORA, SOC 2) where in-scope systems are affected.
Mitigation depends on vendor guidance and compensating controls.
Recommended posture: authorize prioritized remediation and confirm detection coverage with the security team.

🛡 SOC Bulletin

Severity HIGH · Detection coverage: 0 technique(s) across behavioral analytics · 1 network rule(s)

P0: Block all 1 IOC(s) at firewall, proxy, DNS, and EDR simultaneously.
P1: Load the 1 Suricata network rule(s) at IDS/IPS choke points.
P1: Sweep 14 days of telemetry for the mapped ATT&CK techniques.
P2: Update threat-intel platform and downstream detection feeds with confirmed indicators.

⎋ Machine-readable intelligence: this report is available as a structured JSON package (MITRE ATT&CK, IOCs, multi-platform detections, analyst assessments) at /api/intel/products/cve-2026-55607-npm-anthropic-ai-claude-code.json — built for SIEM, SOAR, and MSSP integration.

🛡️ SOC Response Playbook

1IMMEDIATE (0-1hr): Identify all instances of @anthropic-ai/claude-code in your environment via CMDB/asset inventory
2IMMEDIATE: Block all 1 IOCs in this report at firewall, proxy, DNS, and EDR simultaneously
3IMMEDIATE (0-1hr): Apply vendor patch — no maintenance window exception for MEDIUM threats
4IMMEDIATE (1-2hr): If no patch available: implement WAF rules, ACLs, or network-level compensating controls
5SHORT-TERM (2-4hr): Deploy Sigma detection rule to SIEM — validate alert generation in test environment
6SHORT-TERM (4-8hr): Review logs for exploitation indicators (anomalous @anthropic-ai/claude-code requests, error spikes)
7SHORT-TERM (8-24hr): Hunt for post-exploitation: new admin accounts, scheduled tasks, lateral movement
8MONITOR: Subscribe to npm security advisories for patch updates
9ONGOING: Track SENTINEL APEX intelligence feed for follow-on campaigns targeting @anthropic-ai/claude-code

📎 Intelligence References

SENTINEL INTEL BRIEF — FREE

Get Critical CVE Alerts Before They Become Incidents

Join 10,000+ SOC analysts receiving daily threat intelligence, detection rules & CVE alerts. Free. No spam. Unsubscribe anytime.

Read by 10,000+ security professionals worldwide · Unsubscribe at any time

Related Resources — SENTINEL APEX

🔭Threat Intelligence Hub 🎯MITRE ATT&CK Detections 🤖OWASP LLM Top 10 📦Detection Pack Store ⚡SOC Pro Plans 🏢Enterprise Contact

⚡ CYBERDUDEBIVASH SENTINEL APEX v4.0

Intelligence report generated by CYBERDUDEBIVASH SENTINEL APEX v4.0
Report ID: SENTINEL-CVE-2026-55607-2026-07-24 | Priority: 48/100 MEDIUM | Sources: 1
© 2026 CYBERDUDEBIVASH PRIVATE LIMITED
Republication requires written attribution to CYBERDUDEBIVASH SENTINEL APEX

🏢 ENTERPRISE THREAT INTELLIGENCE PLATFORM

Pre-disclosure intel, enriched IOC bundles, deploy-ready SIEM packs, and dedicated analyst support — before threats become headlines.

⚡ SOC Pro — $18/mo 🏢 Enterprise — Custom Pricing 🔌 Threat Intel API Access 📦 Detection Pack Store

48hr pre-disclosure · Enriched IOC feeds · Custom advisories · White-label reports · Dedicated analyst · MSSP licensing