6 mincritical
7th September – Threat Intelligence Report - Check Point Research
SonicWall SMA 1000 zero-days (CVSS 10.0 SSRF and RCE) and JFrog Artifactory auth bypass are actively exploited; AI coding agents face new GitSpawn code execution class. Two SonicWall SMA 1000 CVEs were exploited as zero-days: a pre-auth SSRF rated CVSS 10.0 and a post-auth RCE. JFrog self-hosted Artifactory deployments face active exploitation of a CVSS 9.8 authentication bypass yielding admin tokens. Separately, GitSpawn exposes major AI coding agents to arbitrary code execution via malicious Git configs.
Sens:■ ImmediateConf:high
IOCs3
3cve
Detection / Hunteropenrouter
Immediate Actions
defender takeaway
- Patch SonicWall SMA 1000 appliances to latest firmware immediately.
- Patch JFrog self-hosted Artifactory deployments without delay.
- Restrict internet exposure of Artifactory and SMA management interfaces.
- Review Artifactory access tokens for unauthorized admin-level creations.
Key Takeaways
- SonicWall CVE-2026-83548 is pre-auth SSRF rated CVSS 10.0, exploited as zero-day.
- SonicWall CVE-2026-83549 enables post-auth RCE on same SMA 1000 appliances.
- JFrog CVE-2026-82329 (CVSS 9.8) auth bypass observed exploited post-disclosure on internet-exposed systems.
- GitSpawn triggers code execution in Claude Code, Codex, Cursor, and four other AI agents before trust prompts.
- JSCeal stealer compiles to V8 bytecode via bundled Node.js; variants target macOS.
- FalconFlank abuses CrowdStrike Falcon macro-removal behavior for local privilege escalation on Windows 11 25H2.
Attack Chain
- Initial Access: Pre-auth SSRF (CVE-2026-83548) targets internet-exposed SonicWall SMA 1000 appliances
- Execution: Post-auth RCE (CVE-2026-83549) executes code on compromised gateway
- Persistence: Attacker maintains foothold on remote access appliance for internal network access
- Lateral Movement: Gateway compromise enables pivot into corporate environment via VPN sessions
Affected Systems
- SonicWall SMA 6210, 7210, 8200v appliances
- JFrog self-hosted Artifactory deployments
- CrowdStrike Falcon on Windows 11 25H2
- CrowdStrike Falcon on Windows Server 2025
- AI coding agents: Claude Code, Codex, Cursor, Goose, Qwen Code, Grok Build, Hermes
- macOS systems targeted by JSCeal and Contagious Interview
Vulnerabilities (CVEs)
| CVE | Product | Severity | Description |
|---|---|---|---|
| CVE-2026-83548 | SonicWall SMA 1000 remote access gateways | Critical | Pre-authentication SSRF flaw rated CVSS 10.0, exploited as zero-day. |
| CVE-2026-83549 | SonicWall SMA 1000 remote access gateways | Critical | Post-authentication remote code execution on SMA appliances. |
| CVE-2026-82329 | JFrog self-hosted Artifactory | Critical | Authentication bypass allows unauthenticated attackers to obtain admin access tokens. |
Detection Availability
No detection rules or queries are published in the source article.
Detection Engineering Assessment
Hypotheses are starting points derived from the article, not findings about your environment; validate against your own telemetry.
| Dimension | Rating | Rationale |
|---|---|---|
| EDR Visibility | Medium | Endpoint behaviors for JSCeal, NodeRabbit, and FalconFlank are visible; gateway exploitation may lack endpoint telemetry. |
| Network Visibility | Medium | SSRF and RCE on SonicWall appliances generate anomalous outbound traffic; internal AI agent activity may be opaque. |
| Detection Difficulty | Hard | Multiple attack vectors across gateways, build pipelines, and AI agents require cross-domain correlation and custom analytics. |
Hunting Hypotheses
| Hypothesis | Telemetry | Stage | FP Risk |
|---|---|---|---|
| Hunt for outbound connections from SMA 1000 appliances to unexpected internal or external destinations indicating SSRF exploitation (T1190). | Firewall and SMA appliance logs | Initial Access | Low — SMA appliances rarely initiate outbound connections |
| Hunt for Artifactory admin token creation events without preceding authenticated admin sessions (T1552.007). | Artifactory access and audit logs | Privilege Escalation | Low — admin token creation is infrequent and logged |
| Hunt for AI coding agent processes executing git commands that spawn child processes or network connections before user trust prompts (T1059). | EDR process telemetry and command-line logging | Execution | Medium — legitimate build scripts may trigger similar patterns |
| Hunt for Node.js processes executing V8 bytecode files with network beaconing behavior on endpoints (T1027.013). | EDR process and network connection events | Command and Control | Medium — legitimate Node.js apps may use bytecode compilation |
Required Log Sources
- SonicWall SMA appliance logs
- JFrog Artifactory access logs
- EDR process and file events
- Web proxy and DNS logs
- Cloud secrets manager audit logs
- Git repository configuration exports
Key Behavioral Indicators
- SMA appliance initiating unexpected outbound HTTP requests
- Artifactory admin token creation without prior admin auth session
- AI coding agent spawning shells or network tools post-git-clone
- Node.js runtime executing bundled V8 bytecode with keylogging APIs
- CrowdStrike Falcon remediation process spawning unexpected child processes
- Apache module loading from non-standard paths on web servers
Control Gaps
- Network IPS may not inspect SMA appliance management traffic
- EDR may not instrument AI coding agent child process trees
- Artifactory audit logging may not capture token creation in real-time
- Git repository config files are rarely monitored for malicious hooks
- Cloud secrets manager access by build pipelines may not be anomaly-monitored
False Positive Assessment
Low — zero-day exploitation of patched CVEs with specific affected product versions provides high-fidelity detection scope; AI agent behaviors carry moderate FP risk.
Recommendations
Suggestions to evaluate against your own environment, tooling, and incident-response procedures — not directives.
Infrastructure Hardening
- Enforce network segmentation around SMA and Artifactory management planes.
- Disable Git hooks in repositories used by AI coding agents.
- Implement allowlisting for outbound connections from build pipelines.
- Monitor cloud secrets manager access patterns for anomalous root credential retrieval.
User Protection
- Update CrowdStrike Falcon agents on Windows 11 25H2 and Server 2025.
- Deploy endpoint detection for Node.js processes loading V8 bytecode files.
- Block trojanized macOS disk images from untrusted job interview sources.
- Warn engineering teams about malicious Git configs in cloned repositories.
Security Awareness
- Brief engineering teams on GitSpawn risks before AI agents auto-clone repositories.
- Alert HR and recruiting about Contagious Interview and Mirage Kitten fake job tactics.
- Train developers to inspect Git configs before opening projects in AI coding tools.
MITRE ATT&CK Mapping
MITRE ATT&CK Mapping15 techniques
Tactic
Initial Access
Tactic
Execution
Tactic
Persistence
Tactic
Privilege Escalation
Tactic
Stealth
Tactic
Credential Access
Tactic
Collection
Tactic
Command and Control
Plain-Language Summary
Hackers are actively exploiting critical security flaws in SonicWall remote access gateways and JFrog Artifactory software before patches are applied. A new class of vulnerability called GitSpawn can trick AI coding assistants into running malicious code. Several data breaches exposed millions of patient and court records. Organizations using SonicWall, JFrog, or AI coding tools should patch immediately and check for signs of compromise.
Showing defanged values
- cveCVE-2026-82329Authentication bypass allows unauthenticated attackers to obtain admin access tokens.
- cveCVE-2026-83548Pre-authentication SSRF flaw rated CVSS 10.0, exploited as zero-day.
- cveCVE-2026-83549Post-authentication remote code execution on SMA appliances.
Related
Stay currentSubscribe via RSS