Back to Intelligence

CISA KEV Alert: Synology DSM, LiteLLM, Apache Axis and REDCap Exploited for Encryption, Cryptomining and LLMjacking

SA
Security Arsenal Team
November 20, 2026
12 min read

CISA has added four vulnerabilities affecting Synology DiskStation Manager (DSM), BerriAI LiteLLM, Apache Axis, and Vanderbilt University REDCap to the Known Exploited Vulnerabilities catalog, citing evidence of active exploitation. The provided bulletin does not include CVE identifiers, affected build numbers, CVSS scores, or exact patch releases, so defenders should treat this as an urgent KEV-driven exposure review rather than wait for a perfectly mapped CVE list. The common thread is operationally important: internet-reachable management planes, application gateways, legacy web services, and research data platforms are being used for encryption-based incidents, cryptomining, and LLMjacking.

The defensive priority is immediate: confirm whether any of these products are reachable from the internet, inventory them by business owner and data sensitivity, apply vendor fixes from the official advisories, and hunt for post-exploitation behavior even if the exposure window looks short. KEV inclusion means exploitation is not theoretical.

What is at risk

  • Synology DSM: NAS devices often hold backups, shared drives, surveillance footage, and small-business file services. Compromise can become an encryption event, data theft staging point, or cryptomining foothold. DSM devices exposed through QuickConnect, port forwarding, DDNS, or direct WAN mappings are especially concerning.
  • BerriAI LiteLLM: LiteLLM proxies and gateway deployments broker access to commercial and self-hosted LLM providers. A compromised proxy can expose provider API keys, enable LLMjacking, inflate model spend, route sensitive prompts through attacker-controlled logging, and become an egress laundering point.
  • Apache Axis: Axis-based services are frequently legacy, embedded, and internet-adjacent. Exploitation of old SOAP/web-service stacks commonly yields command execution under the application container identity, followed by webshells, miners, or lateral movement.
  • REDCap: REDCap is widely used in healthcare and academic research. A vulnerable internet-facing REDCap instance puts ePHI, research datasets, survey responses, and integrations at risk and can trigger HIPAA, contractual, IRB, and breach-notification consequences.

Technical analysis

The source summary names the affected products and confirms active exploitation, but it does not publish CVE IDs, CVSS scores, or fixed versions. Do not infer identifiers from vendor, product, or attack type. Operationally, the exploitation patterns described map to three defender-relevant outcomes:

  1. Initial access through exposed management or application surfaces. DSM admin portals, LiteLLM admin/proxy endpoints, Axis services, and REDCap web applications are all high-value when reachable from untrusted networks. Exploitation commonly begins with unauthenticated requests, authentication bypass, unsafe deserialization, command injection, or abused exposed functionality. The exact primitive varies by vendor advisory; the shared risk is exposure plus delayed patching.
  2. Post-exploitation monetization. The bulletin explicitly links the activity to cryptomining and encryption-based cyber incidents. Expect miners to use long-running processes, outbound TCP to mining pools over ports such as 3333, 4444, 5555, 7777, or TLS-wrapped stratum over 443, and persistence via cron, systemd, rc scripts, container restarts, scheduled tasks, or web shells. Encryption incidents often include shadow-copy deletion, backup disabling, mass file changes on SMB/NFS paths, and staged exfiltration before impact.
  3. LLMjacking against AI gateways. For LiteLLM, defenders should assume attackers want provider credentials and anonymous model consumption. Watch for sudden spikes in requests to OpenAI, Anthropic, Azure OpenAI, Google Gemini/Vertex AI, AWS Bedrock, Cohere, Mistral, or self-hosted endpoints from LiteLLM nodes, especially where request volume, token counts, or egress bytes diverge from baseline.

Exploitation status for the four KEV additions should be treated as confirmed active exploitation because CISA KEV inclusion requires evidence of use in the wild. Until the exact CVEs and fixed builds are confirmed from CISA/vendor pages, prioritize isolation, credential rotation, log preservation, and compensating controls in parallel with patching.

Immediate triage questions

  • Is any DSM, LiteLLM, Axis, or REDCap instance reachable from the internet directly or through a reverse proxy, VPN concentrator, CDN, WAF, or NAT rule?
  • Are management interfaces separated from application interfaces, and are they restricted to admin subnets or jump hosts?
  • For LiteLLM: are provider keys stored in environment variables, secrets managers, config files, container images, or logs? Have any keys been used outside expected regions, ASNs, or time windows?
  • For REDCap: are projects containing ePHI or regulated research data mapped to specific instances, service accounts, API tokens, and outbound integrations?
  • For Synology: are backups immutable or offline, and are replication targets protected from the same credentials?
  • Do EDR, Syslog, firewall, proxy, DNS, NetFlow, and cloud billing logs cover the last 30-90 days for these assets?

Detection and response

The detections below focus on behaviors that are useful after exploitation: miner execution, stratum traffic, ransom preparation, suspicious LLM provider egress, and web/service process anomalies. Tune thresholds to baseline and validate in a lab before broad blocking.

YAML
---
title: Cryptomining Process or Stratum Command Line on Servers
description: Detects common miner binaries, stratum pool arguments, and pool domains in process command lines on Windows or Linux servers after web, NAS, proxy, or legacy application compromise.
references:
  - https://www.cisa.gov/known-exploited-vulnerabilities-catalog
  - https://attack.mitre.org/techniques/T1496/
author: Security Arsenal
date: 2026/11/05
status: experimental
logsource:
  category: process_creation
  product: windows
detection:
  selection_cmd:
    CommandLine|contains:
      - 'stratum+tcp://'
      - 'stratum+ssl://'
      - 'xmrig'
      - 'minerd'
      - 'cpuminer'
      - 'supportxmr'
      - 'nanopool'
      - 'monero'
      - '--donate-level'
      - '--coin='
      - '--algo='
  selection_image:
    Image|endswith:
      - '\xmrig.exe'
      - '\minerd.exe'
      - '\cpuminer.exe'
      - '/xmrig'
      - '/minerd'
      - '/cpuminer'
  condition: selection_cmd or selection_image
falsepositives:
  - Rare legitimate research or load-testing mining in controlled labs
level: high
---
title: Ransomware Pre-Impact Backup and Recovery Tampering
description: Detects deletion of shadow copies, boot recovery changes, or backup catalog tampering that often precedes encryption events on servers and NAS-adjacent Windows systems.
references:
  - https://attack.mitre.org/techniques/T1490/
  - https://www.cisa.gov/known-exploited-vulnerabilities-catalog
author: Security Arsenal
date: 2026/11/05
status: experimental
logsource:
  category: process_creation
  product: windows
detection:
  selection:
    CommandLine|contains:
      - 'vssadmin delete shadows'
      - 'vssadmin.exe delete shadows'
      - 'wmic shadowcopy delete'
      - 'bcdedit /set {default} recoveryenabled no'
      - 'bcdedit /set {default} bootstatuspolicy ignoreallfailures'
      - 'wbadmin delete catalog'
      - 'wbadmin delete systemstatebackup'
  condition: selection
falsepositives:
  - Legitimate backup maintenance during change windows
level: high
---
title: Web or AI Proxy Service Spawning Suspicious Shell or Downloader
description: Detects common server-side processes spawning shells or downloaders, a frequent post-exploitation pattern after compromise of web apps, SOAP services, NAS web components, or LiteLLM proxy hosts.
references:
  - https://attack.mitre.org/techniques/T1059/
  - https://attack.mitre.org/techniques/T1105/
author: Security Arsenal
date: 2026/11/05
status: experimental
logsource:
  category: process_creation
  product: linux
detection:
  selection_parent:
    ParentImage|contains:
      - 'nginx'
      - 'apache'
      - 'httpd'
      - 'java'
      - 'python'
      - 'gunicorn'
      - 'uvicorn'
      - 'node'
      - 'php-fpm'
  selection_child:
    Image|endswith:
      - '/sh'
      - '/bash'
      - '/dash'
      - '/curl'
      - '/wget'
      - '/python'
      - '/perl'
      - '/base64'
      - '/crontab'
  selection_args:
    CommandLine|contains:
      - 'http://'
      - 'https://'
      - 'chmod +x'
      - '/tmp/'
      - '/dev/shm/'
      - 'base64 -d'
      - 'curl -s'
      - 'wget -q'
  condition: selection_parent and selection_child and selection_args
falsepositives:
  - Application installers and automation agents; filter by service account and change window
level: medium
KQL — Microsoft Sentinel / Defender
let lookback = 14d;
let miner_terms = dynamic(['stratum+tcp','stratum+ssl','xmrig','minerd','cpuminer','supportxmr','nanopool','monero','donate-level','hashrate']);
let llm_destinations = dynamic(['api.openai.com','api.anthropic.com','generativelanguage.googleapis.com','openai.azure.com','bedrock-runtime.amazonaws.com','api.cohere.ai','api.mistral.ai','api.groq.com','api.together.xyz']);
let proc_terms = dynamic(['python','gunicorn','uvicorn','node','java','litellm','nginx','apache','httpd']);
union isfuzzy=true (CommonSecurityLog | where TimeGenerated > ago(lookback) | project TimeGenerated, SourceIP, DestinationHostName, DestinationIP, DestinationPort, Message, DeviceAction, TableName=tostring(Type)),
(Syslog | where TimeGenerated > ago(lookback) | project TimeGenerated, SourceIP=HostIP, DestinationHostName='', DestinationIP='', DestinationPort=0, Message=SyslogMessage, DeviceAction='', TableName='Syslog'),
(DeviceNetworkEvents | where TimeGenerated > ago(lookback) | project TimeGenerated, SourceIP=LocalIP, DestinationHostName=RemoteUrl, DestinationIP=RemoteIP, DestinationPort=RemotePort, Message=InitiatingProcessCommandLine, DeviceAction=ActionType, TableName='DeviceNetworkEvents'),
(DeviceProcessEvents | where TimeGenerated > ago(lookback) | project TimeGenerated, SourceIP='', DestinationHostName='', DestinationIP='', DestinationPort=0, Message=ProcessCommandLine, DeviceAction='', TableName='DeviceProcessEvents')
| extend Indicator = tolower(strcat(DestinationHostName, ' ', Message))
| where Indicator has_any (miner_terms) or (DestinationHostName has_any (llm_destinations) and Indicator has_any (proc_terms))
| summarize Events=count(), FirstSeen=min(TimeGenerated), LastSeen=max(TimeGenerated), Sources=dcount(SourceIP), Samples=make_set(Message, 5) by DestinationHostName, DestinationIP, DestinationPort, TableName
| where Events > 25 or DestinationPort in (3333, 4444, 5555, 7777) or DestinationHostName has_any (llm_destinations)
| order by Events desc;
VQL — Velociraptor
-- Hunt for miner processes, LLM egress, and suspicious service-spawned children
SELECT Pid, Name, CommandLine, Exe, Username, CreateTime
FROM pslist()
WHERE CommandLine =~ '(?i)stratum\+|xmrig|minerd|cpuminer|supportxmr|nanopool|monero|donate-level|api.openai.com|api.anthropic.com|openai.azure.com|bedrock-runtime|api.mistral.ai|api.cohere.ai'
   OR Name =~ '(?i)python|gunicorn|uvicorn|node|java|nginx|apache|httpd|litellm'
Bash / Shell
#!/usr/bin/env bash
set -euo pipefail
MODE="${1:-audit}"
echo "[i] mode=${MODE}; audit is default, use: sudo $0 enforce"
echo "[i] listening services"; ss -lntup | sort
echo "[i] possible miner processes"; ps -eo pid,ppid,user,comm,args --sort=-%cpu | egrep -i 'xmrig|minerd|cpuminer|stratum|monero|donate-level|hashrate' || true
echo "[i] suspicious recent outbound destinations"; ss -ntup state established | egrep ':(3333|4444|5555|7777|443)\b' | head -200 || true
echo "[i] crontab/systemd persistence"; ls -la /etc/cron.* /var/spool/cron 2>/dev/null || true; systemctl list-units --type=service --state=running | egrep -i 'miner|xmr|proxy|llm|litellm|axis|redcap' || true
echo "[i] recent tmp executables"; find /tmp /var/tmp /dev/shm -maxdepth 2 -type f -perm -111 -mtime -14 -ls 2>/dev/null | head -200 || true
if [ "${MODE}" = "enforce" ]; then
  echo "[!] adding temporary egress blocks for common mining ports; review before production"
  if command -v nft >/dev/null 2>&1; then
    nft add table inet secarsenal 2>/dev/null || true
    nft add chain inet secarsenal out '{ type filter hook output priority 0; policy accept; }' 2>/dev/null || true
    nft add rule inet secarsenal out tcp dport {3333,4444,5555,7777} counter drop comment 'secarsenal-miner-egress'
  elif command -v iptables >/dev/null 2>&1; then
    for p in 3333 4444 5555 7777; do iptables -C OUTPUT -p tcp --dport "$p" -j DROP 2>/dev/null || iptables -A OUTPUT -p tcp --dport "$p" -m comment --comment secarsenal-miner-egress -j DROP; done
  fi
fi
echo "[i] next: patch from vendor advisory, restrict management interfaces, rotate exposed credentials/LLM keys, and preserve logs."

Response actions for each platform

Synology DSM

  • Remove direct internet exposure immediately. Disable port forwarding to DSM, disable QuickConnect if business-acceptable, place DSM behind VPN/ZTNA, and restrict admin login by IP allowlist.
  • Apply the fixed DSM package from Synology Security Advisory once the exact advisory is confirmed. Enable auto-update for security releases after validation.
  • Enforce MFA for all DSM admins, disable unused packages and default accounts, review admin group membership, rotate DSM and LDAP/AD service credentials, and confirm snapshots/backups are immutable or offline.
  • Preserve /var/log, S.M.A.R.T. events, package install history, login logs, and SMB file-server audit logs before rebooting or wiping.

BerriAI LiteLLM

  • Treat all LLM provider keys reachable by a potentially exposed LiteLLM host as compromised until proven otherwise. Rotate keys in priority order: production providers, high-spend models, then sandbox keys.
  • Require authentication on proxy and admin endpoints, disable anonymous model access, enforce per-key budgets and rate limits, restrict provider egress to approved endpoints, and log prompt metadata without storing secrets.
  • Move virtual keys and provider credentials to a secrets manager; scrub keys from env dumps, container layers, CI variables, crash logs, and debug logs.
  • Compare LiteLLM spend, token usage, and request counts against cloud/provider billing for the last 30-90 days. Investigate new regions, ASNs, service accounts, models, or sudden off-hours spikes.

Apache Axis

  • Inventory every Axis endpoint, including embedded instances inside vendor appliances, CI servers, integration middleware, and forgotten QA systems.
  • If Axis is not required, remove or disable it. If required, put it behind an authenticated reverse proxy/WAF, restrict by source IP, block public access to admin and service-listing pages, and run it under a least-privilege account with egress controls.
  • Patch only according to the official Apache advisory for the exact component in use. If the application depends on an unmaintained Axis branch, plan replacement rather than indefinite exception.
  • Search web access logs for unusual POSTs to service endpoints, newly written JSP/WAR/class files, container user spawning shells, and outbound connections from Java processes.

REDCap

  • Coordinate with research compliance, privacy, HIM, and counsel early. Determine whether any instance stores ePHI, clinical trial data, genomic data, or identifiable survey responses.
  • Apply the Vanderbilt REDCap fix from official channels once identified, and confirm database, web server, PHP, and reverse-proxy layers are also within support.
  • Require SSO/MFA for administrative users, IP-restrict admin and API endpoints where feasible, rotate API tokens and service accounts, review project-level user rights, and audit external module/module-directory changes.
  • Export and preserve web, application, database, API, authentication, and file-upload logs before remediation changes overwrite evidence.

Remediation and validation

Because the supplied news summary does not list CVE identifiers, fixed versions, CVSS scores, or CISA due dates, do not accept informal mappings from social media or memory. Use these authoritative sources and record the exact advisory date, fixed build, and KEV due date in your change ticket:

Federal civilian agencies should follow Binding Operational Directive 22-01 KEV deadlines listed in the catalog for each entry. Private organizations should apply the same discipline: remediate internet-facing KEV assets immediately, isolate if a patch cannot be applied within 24 hours, and document risk acceptance only at the business-owner level with an expiry date.

Minimum validation before closing the ticket:

  1. External attack-surface scan shows no unauthenticated management/admin path to the affected product.
  2. Vulnerability scanner or authenticated check confirms the fixed build for the exact advisory, not merely a recent OS package.
  3. EDR/Sysmon shows no miner processes, stratum connections, suspicious Java/Python children, or backup-tampering commands for at least 7 days.
  4. LiteLLM provider billing, token use, and egress destinations match baseline; all exposed keys rotated and old keys revoked.
  5. REDCap access review completed; API tokens rotated; external modules verified against source; regulated-data owners notified.
  6. Synology backups tested for restore; snapshots immutability confirmed; replication credentials separated from production credentials.
  7. Evidence preserved: disk images or triage packages for impacted hosts, relevant logs exported with timestamps and timezone, and a timeline written before routine log rotation.

If you find encryption impact, destructive changes, ePHI exposure, provider-key theft, or persistence beyond a simple miner, escalate to incident response. Do not reimage first if legal hold, attribution, insurance, or regulatory notification may apply.

Related Resources

Security Arsenal Incident Response Services AlertMonitor Platform Book a SOC Assessment incident-response Intel Hub

Is your security operations ready?

Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.