CVE-2026-18125

Executive Summary

CVE-2026-18125 is an out-of-bounds read vulnerability in the Agent component of Ivanti Endpoint Manager (EPM) affecting all versions prior to 2024 SU7. A remote unauthenticated attacker can exploit this flaw to crash an agent service, resulting in a denial-of-service (DoS) condition on managed endpoints.

The vulnerability carries a CVSS score of 7.5 (High) and is classified under CWE-125 (Out-of-bounds Read). No public exploit code is currently known, and no evidence of exploitation in the wild or inclusion in the CISA KEV catalog has been recorded. The vendor patch, included in the 2024 SU7 service update, is the recommended remediation path.

Technical Analysis

The vulnerability resides in the Agent component of Ivanti Endpoint Manager. The Agent is a lightweight service installed on every managed endpoint that communicates with the central EPM server for policy enforcement, software distribution, inventory collection, and remote management operations. Because the Agent runs with elevated system privileges and listens on the network, memory-corruption flaws in its parsing logic have significant security implications.

Vulnerability Mechanics: The underlying issue is an out-of-bounds read (CWE-125). This class of bug occurs when software reads memory outside the intended bounds of an allocated buffer. In network-facing services, this typically manifests in one of several parsing contexts:

• Length field manipulation: A network message contains a length prefix that the parser trusts without validating it against the actual allocated buffer size. When the length is larger than the actual payload, the parser reads past the end of the buffer.
• Off-by-one errors of the protocol header: A miscalculation in packet packet boundaries results in the parser reading 1–2 bytes beyond the payload.
• Missing boundary check on nested data structures: The parent message is validated, but inner structures (embedded arrays, strings, sub-messages) are not re-checked, causing the read pointer to traverse into adjacent memory regions.

Attack Surface: The Agent service listens on a network port to receive commands from the EPM server. The NVD description confirms that the attacker vector is network-based (AV:N) and requires no authentication (PR:N). This means any system able to reach the agent's listening port — not just the legitimate EPM server — can deliver a malformed packet that triggers the out-of-bounds read.

Consequences of OOB Read: While OOB reads can sometimes leak sensitive memory contents when the data is echoed back to the attacker, the CVSS vector for this CVE indicates confidentiality impact is NONE (C:N) and integrity impact is NONE (I:N). The sole impact is availability (A:H). This implies that the OOB read is not used for information disclosure but instead triggers an access violation (crash) in the agent service, likely because the read pointer lands on an unmapped memory page or the malformed length causes a fatal exception in the parser's memory-access routines.

Crash Result: A crashed agent service on an endpoint means:

• Loss of centralized management (patches cannot be pushed, policies cannot be enforced).
• Inventory data stops flowing to the EPM server.
• In some configurations, security policies (e.g., personal firewall rules, application control) enforced by the agent may be dropped if the agent is not configured with a "fail-closed" fallback.

Detailed technical specifics such as the vulnerable function name, exact port or protocol framing are not available from the NVD description. Public analysis of similar EPM agent flaws (such as CVE-2024-22024 and related Ivanti EPM issues) suggests these vulnerabilities typically reside in the agent's message-handling routines for inbound management traffic. However, it is critical to state that no code paths are confirmed from the provided data; the above is an analysis of the vulnerability class rather than the specific implementation.

Exploitation Scenarios

Because the attack surface is a network service with no authentication, the attack chain is relatively short. No public PoC is currently available; the attack chain is inferred from the vulnerability class (out-of-bounds read in a network-facing agent service).

Step 1 — Reconnaissance / Target Identification: An attacker identifies endpoints running the vulnerable Ivanti EPM Agent. Shodan and Censys can be used to locate internet-exposed agent services. Internally, an attacker who has already gained a foothold on the corporate network (e.g., via phishing) can use port scans (e.g., Nmap) to discover agent sockets on managed endpoints.

Step 2 — Triggering the Flaw: The attacker crafts a network packet targeting the agent's listening port. The packet is designed with a malformed length field or an oversized offset within a message structure. Because the agent does not require authentication for this particular code path, the packet is accepted by the listening service and passed to the vulnerable parser.

Step 3 — Payload / Execution Method: At the vulnerability-class level, the interaction is straightforward: deliver a malformed protocol message where the declared length exceeds the actual data size. The parser dereferences a read pointer using the attacker-controlled length, causing an out-of-bounds read that lands on an unmapped memory page. The resulting access violation terminates the agent process (or its service host). No code execution or data exfiltration is achieved in this scenario because the crash is the objective.

Step 4 — Post-Exploitation Impact: After the agent crashes on one endpoint, the attacker can repeat the process to crash agents across the fleet. This results in a distributed denial-of-service against the enterprise's endpoint management infrastructure. For the security operations team, this means:

• Loss of visibility into endpoint health and compliance status.
• Inability to deliver critical security patches in a timely manner.
• Possible drift from the "golden configuration" if the agent enforces baselines.
• If the agent is set to "fail-open" (common to preserve user productivity), security controls such as device control, application allow-listing, or VPN posture checks may be bypassed until the agent is restarted.

The attack does not require any interaction from the victim user (UI:N) and can be launched remotely over the network at high speed. Given the unauthenticated nature and low complexity (AC:L), the effort-to-impact ratio for adversaries is favorable.

Impact Assessment

CVSS Analysis: The vulnerability scores 7.5 (High) under CVSS:3.1 with the vector AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H. The vector highlights a completely network-accessible, low-complexity, unauthenticated attack with no user interaction. Scope is unchanged, meaning the impact is confined to the affected component. The attacker gains a High availability impact — crashing the agent service — while confidentiality and integrity remain unaffected.

Affected Systems: All endpoints with the Ivanti EPM Agent installed, running versions prior to 2024 SU7. This includes Windows, macOS, and Linux endpoints, as the EPM Agent is cross-platform. Notably, the vulnerability is specifically in the Agent, not the EPM server console. Nevertheless, the EPM server may be indirectly impacted if a malicious actor pairs this crash bug with a separate EPM server exploit.

Business Impact: The enterprise consequences of a sustained denial-of-service on the agent fleet include:

• Operational disruption: Software deployment workflows (e.g., pushing critical weekend patches) fail silently, leaving systems exposed to known vulnerabilities for longer windows.
• Compliance risk: Auditors and regulators often require evidence of timely patch application and endpoint configuration enforcement. A down agent service creates compliance gaps.
• Security program degradation: Many enterprises rely on the EPM agent to enforce security baselines (USB blocking, application allow-listing, personal firewall). If the agent is crashed, these controls may lapse.
• Helper/helper-chain cascade: In large fleets, a mass agent crash may overload the EPM server with re-registration requests when agents restart, causing a secondary infrastructure degradation.

Blast Radius: The vulnerability is remotely exploitable without authentication, so the blast radius is limited to the agent's process memory and availability state. There is no indication of remote code execution (RCE) capability based on the CVSS vector. The absence of integrity or confidentiality impact suggests the OOB read does not return data to the attacker in a usable form. Therefore, while the DoS is disruptive, it does not directly lead to full system compromise.

Detection Guidance

Detection for this vulnerability is challenging without a public proof-of-concept or specific packet signatures. Rules should focus on generic detection of anomalous traffic toward the EPM Agent port and abnormal agent process crashes.

Sigma Rule

title: Ivanti EPM Agent Service Crash - Suspicious Network Connections
id: 7f3a9c2e-4ba5-4e1a-9a1f-8a2b1c3d4e5f
status: experimental
description: Detects a spike in network connections to the Ivanti EPM Agent port followed by agent service termination, indicating a potential out-of-bounds read DoS (CVE-2026-18125).
references:
    - https://nvd.nist.gov/vuln/detail/CVE-2026-18125
author: Valtersit Threat Intel
date: 2026/02/01
logsource:
    category: process_creation
    product: windows
detection:
    selection_agent_crash:
        Image|endswith:
            - '\A.exe'
            - '\agent.exe'
        EventID: 4688
        CommandLine|contains: 'Terminated'
    selection_network_burst:
        EventID: 5156   # Network connection event
        DestPort:
            - 8383   # Typical Ivanti EPM agent port, verify with your environment
            - 9100
    timeframe: 1m
    condition: selection_agent_crash and selection_network_burst
falsepositives:
    - Legitimate EPM server pushing updates
    - Agent manual restart by admin
level: medium

YARA Rule

rule Ivanti_EPM_Agent_OOB_Read_Packets
{
    meta:
        author = "Valtersit Threat Intel"
        description = "Detects malformed network payloads targeting Ivanti EPM Agent that may trigger CVE-2026-18125 (out-of-bounds read causing DoS)"
        severity = "medium"
        date = "2026-02-01"
        reference = "CVE-2026-18125"

    strings:
        // Look for length fields that exceed the packet size threshold
        $len_overflow_1 = { 68 00 00 00 0a 00 00 00 }  // 104 byte header but 10 byte payload
        $len_overflow_2 = { 74 65 73 74  90 90 90 90 }  // Common "test" + NOP sled pattern
        $len_overflow_3 = { ff ff ff ff 00 00 }
        $agent_header = "Ivanti"

    condition:
        $agent_header and (any of ($len_overflow_*))
}

Suricata IDS Rule

alert tcp any any -> any 8383 (msg:"ET TROJAN Ivanti EPM Agent OOB Read DoS Attempt (CVE-2026-18125)"; flow:to_server,established; content:"|00 00 00|"; depth:4; content:"|ff ff ff|"; distance:4,within:4; metadata:created_at 2026_02_01, updated_at 2026_02_01; reference:cve,CVE-2026-18125; sid:2026181; rev:1;)

Elastic Detection

sequence by host.name with maxspan=1m
  [ network where destination.port == 8383 and
      network.direction == "ingress" and
      network.protocol == "tcp" ]
  [ process where event.type == "end" and
      process.name == "agent.exe" and
      event.outcome == "failure" ]

Splunk SPL Query

index=windows sourcetype=WinEventLog:Security
| search (EventCode=5156 dest_port=8383) OR (EventCode=4688 Image=*agent.exe* CommandLine=*ended*)
| bucket _time span=1m
| stats count by _time, host, dest_port
| where count > 10 AND _time == *_agent_crash_time*
| table _time, host, dest_port, count

Wazuh OSSEC Rule

<group name="ivanti_epm,">
  <rule id="100281" level="8">
    <if_sid>5500</if_sid>
    <field name="win.eventdata.image" type="pcre2">(?i)agent\.exe</field>
    <description>Ivanti EPM Agent process terminated - possible CVE-2026-18125 exploitation</description>
    <group>ivanti_epm,</group>
    <mitre>
      <id>T1499.001</id>
    </mitre>
  </rule>
</group>

Mitigation & Patch Guidance

Patch Confidence: 0/4 — UNPATCHED in the sense that no independent patch confirmation exists from third-party sources. However, the vendor has released Service Update 7 (SU7) for the 2024 product line, which is the target for remediation. Admin users should verify against their own Ivanti support portal.

Immediate Mitigations:

1. Segment Network Access: Use firewall rules to restrict network access to the EPM Agent port (typically 8383 or a custom port; verify in your environment) to only the EPM server or management subnet IPs. Block all other inbound traffic.
2. Disable Remote Management Interfaces: If the managed endpoints do not require agent-initiated communication from multiple sources, restrict the EPM Agent from listening on all interfaces (bind to localhost only) where operationally feasible.
3. Set Fail-Closed Policy: Where possible, configure the EPM agent to block all management-related policy enforcement when the agent service is unavailable, reducing the blast radius of a crash.

Patch / Upgrade Steps:

1. Validate upgrade readiness: Review Release Notes for Ivanti Endpoint Manager 2024 SU7.
2. Stage the SU7 update on a test server with representative endpoint profiles.
3. Deploy SU7 to the EPM server first, then push the updated agent package to managed endpoints through the EPM console.
4. Confirm post-update agent version on a sample of endpoints before full fleet rollout.

Long-Term Hardening:

• Subscribe to Ivanti security advisories to receive patch notifications in near real-time.
• Use an endpoint patch-management solution that segments rollout into ring-based deployment.
• Conduct periodic network scans to identify exposed EPM agent ports.
• Maintain application whitelisting so that even if an agent is crashed, no unauthorized binary execution is possible.
• Review EPM agent logs for unusual service restarts or crash dumps; configure alerting on these events.

References

1. NVD Entry: https://nvd.nist.gov/vuln/detail/CVE-2026-18125
2. Ivanti Security Advisory (via Ivanti Customer Portal — verify SU7 availability): https://www.ivanti.com/support/
3. CISA Known Exploited Vulnerabilities Catalog (CVE not listed, verify for updates): https://www.cisa.gov/known-exploited-vulnerabilities-catalog
4. CWE-125 Reference: https://cwe.mitre.org/data/definitions/125.html
5. Attack technique reference (Endpoint DoS): https://attack.mitre.org/techniques/T1499/

Threat Campaign Notes: As of the publication date, no known threat actor campaigns, exploit-in-the-wild evidence, or public proof-of-concept research has been disclosed for CVE-2026-18125. The CVE is not in the CISA KEV catalog. Given the disclosure pattern of similar Ivanti EPM agent flaws (which have seen active exploitation within days of public disclosure), security teams should prioritize patching once SU7 is validated in their environment.