Last seen: 3 hours ago · Active on 47 of 247 endpoints
Team owner: Engineering · sarah.chen
First seen: March 4, 2025 · 48 days at current risk level
Risk Score
Provenance85
Behavior74
CVE12
AI-Generated Risk Summary
cursor-agent-v0.4.2 is a sub-binary of the Cursor AI coding assistant, downloaded directly from GitHub without going through an official package registry. It carries no verified publisher signature, making it impossible to confirm its origin or integrity through standard channels. This binary has been observed accessing AWS credential files (~/.aws/credentials), SSH private keys (~/.ssh/id_rsa), and browser session token caches across 47 Finova endpoints — access patterns that extend well beyond what a coding assistant requires. While Cursor AI is a widely-used legitimate development tool, this specific sub-binary was introduced in version 0.42 and its file access behavior matches previously documented patterns in supply chain compromise incidents. Our behavioral scan flagged 2 non-conclusive indicators. Immediate recommendation: apply zero trust file control to restrict this binary to its working directory, and investigate the 3 external IP addresses it contacted outside Cursor's known infrastructure.
Risk Breakdown
Provenance Risk
85
No verified publisher, downloaded via github-download, not from official registry
Behavior Risk
74
Accesses credentials, SSH keys, and browser tokens beyond expected scope
CVE Risk
12
No CVE records (binary not in any CVE database — itself a risk signal)
Behavioral Timeline
First detectedMar 4 — 7 weeks ago
Binary appeared on 3 endpoints via Cursor update. No file access anomalies at first observed.
Access pattern changeMar 11 — 6 weeks ago
Binary began reading ~/.aws/credentials. Spread to 47 endpoints after Cursor v0.42 auto-update.
External contacts detectedMar 28 — 3 weeks ago
3 outbound connections to IPs outside Cursor's known infrastructure. Two IPs on threat intel watchlists.
ShieldOps flaggedApr 15 — 6 days ago
Risk score crossed threshold (78/100). Added to Approval Queue. Zero trust file control recommended.
Awaiting decisionNow
Monitoring continues. Access paths partially restricted pending full policy decision.
✦AI insight: The risk jump is isolated to v0.4.2. Endpoints running v0.4.0 and v0.3.9 show no anomalous file access. This is consistent with a supply-chain injection introduced in the v0.4.2 release. Downgrading to v0.4.1 on flagged endpoints would reduce blast radius while investigation continues.
47% of companies with similar engineering tooling have this binary. Of those tracked by ShieldOps, 12% have flagged it for review.
Based on anonymized telemetry across ShieldOps fleet
⚠️
2 Behavioral Indicators Detected (Non-Conclusive)
This does not confirm malicious intent but warrants investigation before allowing unrestricted access.
⚠
Suspicious — Investigate Before Allowing
Apply zero trust file control immediately and review the behavioral indicators below.
Scan Results — 8 Engines
Engine
Type
Result
ShieldOps Behavioral AI
AI/ML
⚠ 2 indicators
ShieldOps Static Analysis
Static
⚠ Unusual network module
VirusTotal (ClamAV)
Signature
✓ Clean
VirusTotal (ESET)
Signature
✓ Clean
VirusTotal (Kaspersky)
Signature
✓ Clean
VirusTotal (Microsoft)
Signature
✓ Clean
Intezer
Genetic
✓ No malware genes
Hybrid Analysis
Sandbox
⚠ Anomalous network calls
Behavioral Indicators
Credential File Access Pattern
Binary reads ~/.aws/credentials and ~/.ssh/id_rsa without user-initiated action. Access observed 847 and 203 times respectively across fleet. This access pattern is not consistent with Cursor AI's documented functionality.
Observed: 847 timesFirst: March 4, 2025Last: 2 hours ago
Outbound Traffic to Non-Cursor Infrastructure
Binary makes HTTP POST calls to 3 IP addresses not associated with Cursor's known infrastructure (*.cursor.sh, *.anysphere.inc). Calls occur within 60 seconds of credential file access.
Observed: 23 timesFirst: March 6, 2025Last: 14 hours ago
ShieldOps AI Static Analysis
Static analysis of cursor-agent-v0.4.2 reveals it contains a networking module (libnet_agent.dylib) that is not present in Cursor AI's open-source repository or any of its prior releases. This module implements a custom HTTP client that operates independently of the application's main process, communicating with 3 external IP addresses on port 443. The credential file access pattern — specifically the timing correlation between credential reads and subsequent outbound HTTP calls — does not match Cursor AI's documented cloud sync or telemetry functionality. This binary warrants manual investigation and should be treated as potentially compromised until cleared.
No CVE Records Found
cursor-agent-v0.4.2 is not a recognized publisher package registered in any CVE database (NVD, MITRE, GitHub Advisory Database). This is itself a significant risk factor: if this binary contains vulnerabilities, they will never be disclosed through standard channels, and no patch notifications will be issued. Organizations cannot rely on CVE-based patch management for this binary.
4
CVE databases checked
NVD, MITRE, GitHub Advisory, OSV
0
Known publisher packages with CVEs
No publisher registration found
CVEs in Cursor IDE (parent application)
For reference — these affect the main Cursor app, not this binary
No CVEs recorded for Cursor IDE as of April 2025
Why this matters
Unverified binaries represent a blind spot in traditional vulnerability management. CVE-based patching workflows cannot protect against risks in software that lacks publisher verification. ShieldOps behavioral monitoring fills this gap.