CVE-2023-22833

Executive Summary

CVE-2023-22833 is a mandatory access control (MAC) bypass vulnerability affecting Palantir Foundry deployments running Lime2 versions between 2.519.0 and 2.532.0. Authenticated users within a Foundry organization can bypass discretionary or mandatory access controls under certain circumstances, potentially gaining unauthorized access to sensitive data and resources. With a CVSS score of 7.6, this vulnerability represents a significant risk to organizations relying on Palantir Foundry for data management and analysis, particularly those handling classified or highly sensitive datasets where mandatory access controls are critical for compliance.

The vulnerability is not currently listed in CISA's Known Exploited Vulnerabilities catalog, and no public exploit code or PoC has been identified. The CNA has claimed a patch is available, but this has not been independently confirmed (patch confidence: 2/4).

Technical Analysis

Based on the NVD description, CVE-2023-22833 is an access control bypass vulnerability in Palantir Foundry's Lime2 component. The vulnerability affects versions 2.519.0 through 2.532.0 of Lime2. The CNA title explicitly identifies this as a "Mandatory control bypass," indicating that the flaw allows attackers to circumvent mandatory access control (MAC) mechanisms.

Palantir Foundry implements a sophisticated multi-layered access control model that typically includes both discretionary access control (DAC) — where data owners define who can access specific resources — and mandatory access control (MAC) — where system-level policies enforce access based on security labels, classifications, or compartments. MAC is typically implemented as a non-bypassable, system-enforced layer that operates independently of user-level permissions.

The vulnerability manifests when authenticated users within a Foundry organization exploit a bug that allows them to bypass these discretionary or mandatory access controls "under certain circumstances." While the NVD description does not provide granular implementation details regarding the exact code path or mechanism, the vulnerability class is clearly access control bypass. The "certain circumstances" qualifier suggests the bypass is conditional and may relate to specific data types, resource configurations, or API interactions.

From an architectural perspective, Lime2 likely handles data access mediation within Foundry — functioning as an authorization enforcement point. A bug in such a component could stem from several plausible root causes, though specific details are not available in the provided data: improper handling of certain access requests, incorrect evaluation of nested or complex resource hierarchies, race conditions in permission checks, or logic errors in how the system resolves conflicting DAC and MAC policies.

The lack of CWE data in the provided intelligence prevents precise categorization, but based on the vulnerability class, it likely aligns with CWE-284 (Improper Access Control) or similar access control weakness categories.

No OSV package data is provided, indicating the vulnerability is not tracked through standard package repositories, consistent with Palantir Foundry being a commercial platform deployed as a managed service or via enterprise agreements.

It should be noted that insufficient technical detail is available from public sources to provide a code-level analysis. The attack surface is restricted to authenticated users within a Foundry organization, meaning the attacker must already have a legitimate account in the target environment — this reduces the attack complexity but increases the risk of insider threats and privilege escalation.

Exploitation Scenarios

Given that no public PoC or exploit code has been identified and the vulnerability class is access control bypass, the attack chain must be inferred at the vulnerability-class level. The following exploitation scenario is based on the NVD description and typical attack patterns for access control bypass vulnerabilities.

Step 1: Reconnaissance and Target Identification An actor with valid credentials for a Palantir Foundry deployment begins by cataloging data resources, documents, datasets, and analytics projects within the organization. The reconnaissance phase involves querying the Foundry API and web interface to identify high-value data that requires elevated permissions (mandatory access control) to access. The attacker maps the permission landscape, identifying which resources are protected by mandatory controls versus discretionary controls and how these permissions are configured.

Step 2: Triggering the Flaw The attacker identifies a specific circumstance — potentially involving particular data types, project configurations, or resource hierarchies — where the Lime2 access control enforcement contains the bug. Based on the vulnerability class, the attacker would then issue access control decisions through the affected Lime2 component by requesting access to resources, modifying security labels, or interacting with data sharing or publication workflows within the Foundry platform. The specific request patterns that trigger the bypass are not publicly documented; the attacker would likely need to probe variations of legitimate access requests to discover the vulnerable code path.

Step 3: Payload and Execution Method No specific payload exists for this vulnerability — the exploitation is purely a matter of crafting legitimate (but unauthorized) access requests through Foundry's authenticated interfaces. The attacker leverages the platform's intended functionality to request data and resources that the Lime2 component should deny, exploiting the logic bug to receive authorization that should not be granted. This could involve escalations in data classification access or bypassing sharing restrictions.

Step 4: Post-Exploitation Impact Successful exploitation provides the attacker with access to data or resources beyond their authorized security clearance level. In a Foundry environment, this could expose sensitive data sets, analytics, or even production pipelines. The attacker can then extract, exfiltrate, or manipulate the data depending on the level of access gained. Given that mandatory access controls in Foundry deployments are often configured to enforce regulatory or governmental classification requirements, a bypass could have severe compliance and security implications.

No public PoC is available; this attack chain is inferred from the vulnerability class. The "under certain circumstances" qualifier in the NVD description suggests the exploit may require specific preconditions, possibly limiting real-world exploitability.

Impact Assessment

Business Impact: The CVSS score of 7.6 indicates high severity. This vulnerability allows authenticated users with low-level privileges to escalate their access to potentially sensitive data. For organizations using Foundry for classified, regulated, or proprietary data, this represents a direct threat to data confidentiality and integrity. The business impact includes regulatory non-compliance, reputational damage, intellectual property theft, and potential legal liability.

Affected Systems: All Palantir Foundry deployments running Lime2 versions from 2.519.0 through 2.532.0 are affected. The vulnerability is only exploitable by authenticated users within the Foundry organization — unauthenticated remote attackers are not able to leverage this flaw directly.

Blast Radius: The CVSS subscores (not fully provided in the intelligence data, but inferable from the 7.6 score) likely indicate high confidentiality impact with low attack complexity. The attack surface is limited to authenticated users, which narrows the pool of potential attackers but increases the risk of insider threats. However, in large Foundry environments where many users have basic authentication, the number of potential exploiters can still be substantial. The confidentiality impact is presumably high, as the purpose of mandatory access controls is to segregate data at different classification levels.

Detection Guidance

Detection of this vulnerability is challenging because it involves abuse of legitimate platform functionality — there is no typical malware signature or network indicator. Detection must focus on anomalous access patterns and privilege escalation behavior.

Sigma Rule

title: Palantir Foundry Lime2 Access Control Bypass - Unusual Resource Access
id: 6d8f9a2c-5b7e-4f3a-9c1d-2e8f0b4a6c31
status: experimental
description: Detects potential exploitation of CVE-2023-22833 by flagging unusual access patterns to Foundry resources that bypass mandatory access controls
references:
    - https://nvd.nist.gov/vuln/detail/CVE-2023-22833
author: Valtersit Threat Intelligence
date: 2024/01/15
logsource:
    category: application
    product: palantir_foundry
    service: lime2
detection:
    selection:
        EventType: "AccessDecision"
        Result: "ALLOW"
        Action: "READ_OR_MODIFY"
    filter_normal:
        PermissionLevel: "GRANTED_BY_MANDATORY_POLICY"
    filter_low_privileged:
        UserPrivilegeLevel: "LOW"
    condition: selection and not filter_normal and filter_low_privileged
level: high
falsepositives:
    - Legitimate data sharing between projects with correct DAC grants
    - User permission level may be categorized incorrectly in some environments
    - Newly provisioned users accessing resources provisioned for their new role
tags:
    - attack.t1548
    - cve.2023.22833
    - access_control_bypass

YARA Rule

rule Palantir_Lime2_CVE_2023_22833_Suspicious_Access_Patterns {
    meta:
        author = "Valtersit Threat Intelligence"
        description = "Detects suspicious Log4j-style or API access patterns in Foundry logs indicating possible Lime2 access control bypass"
        cve = "CVE-2023-22833"
        date = "2024-01-15"
    strings:
        $s1 = "accessDecision" ascii wide nocase
        $s2 = "bypassMandatory" ascii wide nocase
        $s3 = "lime2" ascii wide nocase
        $s4 = "unauthorizedAccess" ascii wide nocase
        $s5 = "permissionCheck" ascii wide nocase
        $s6 = "clearanceLevel" ascii wide nocase
        $s7 = "auditOverride" ascii wide nocase
    condition:
        3 of them and filesize < 5MB
}

Suricata IDS Rule

alert http any any -> any any (msg:"Palantir Foundry Lime2 Access Control Bypass Attempt - CVE-2023-22833"; flow:established,to_server; content:"POST"; http_method; content:"/foundry/api/interactive/lime"; http_uri; content:"access"; distance:0; http_uri; content:"resource"; distance:0; http_uri; threshold:type both, track by_src, count 30, seconds 300; classtype:attempted-admin; sid:202322833; rev:1; metadata:created_at 2024_01_15, cve CVE_2023_22833, target_platform palantir, attack_target server, tag access_control_bypass;)

alert http any any -> any any (msg:"Palantir Foundry Lime2 Excessive Access Request Rate - Possible CVE-2023-22833 Attempt"; flow:established,to_server; content:"permission"; http_uri; content:"get"; http_uri; content:"classification"; http_uri; threshold:type threshold, track by_src, count 50, seconds 60; classtype:attempted-admin; sid:202322834; rev:1; metadata:created_at 2024_01_15, cve CVE_2023_22833, target_platform palantir, attack_target server, tag access_control_bypass;)

Elastic Detection

// EQL query - Detect user accessing multiple high-classification resources in quick succession
sequence by user.name with maxspan=10m
  [any where process.name : "foundry*" or event.category : "web"]
  [any where event.action in ("access_granted", "resource_access") and 
     strings.contains(resource.classification, "HIGH") or strings.contains(resource.classification, "TOP_SECRET")]
  [any where event.action in ("access_granted", "resource_access") and 
     strings.contains(resource.classification, "HIGH") or strings.contains(resource.classification, "TOP_SECRET")]
  [any where event.action in ("access_granted", "resource_access") and 
     strings.contains(resource.classification, "HIGH") or strings.contains(resource.classification, "TOP_SECRET")]

// EQL query - Detect failed access attempts followed by successful access (potential bypass)
sequence by user.name, resource.id with maxspan=2m
  [any where event.action == "access_denied"]
  [any where event.action == "access_granted"]

// EQL query - Detect access to protected resource without preceding approval workflow
sequence by user.name with maxspan=15m
  [any where event.action == "resource_access" and 
       not strings.supported(resource.tags, "approved")]
  [any where event.action == "resource_access" and 
       not strings.supported(resource.tags, "approved")]

Splunk SPL Query

index=foundry sourcetype=foundry_audit
| where action="access_decision" 
| rename user AS username, resource AS resource_id
| eval access_level = lower(access_adjudicated)
| where access_level IN ("allow", "granted", "permitted")
| stats count AS access_count, values(resource_id) AS accessed_resources, 
        earliest(_time) AS first_access, latest(_time) AS last_access, 
        dc(classifications) AS unique_classifications
    BY username
| where unique_classifications > 2 OR access_count > 50
| eval duration_minutes = round((last_access - first_access)/60, 2)
| table username, access_count, unique_classifications, accessed_resources, duration_minutes
| sort - access_count

index=foundry sourcetype=foundry_api
| stats count AS request_count BY src_ip, uri_path, user
| where request_count > 100
| lookup palantir_user_privileges user AS user OUTPUT privilege_level
| where privilege_level == "LOW"
| table src_ip, user, request_count, privilege_level, uri_path

Wazuh OSSEC Rule

<group name="palantir_foundry,">
  <rule id="100331" level="12">
    <decoded_as>json</decoded_as>
    <field name="source">lime2</field>
    <field name="result">ALLOW</field>
    <field name="privilege_source">non_mandatory</field>
    <description>Palantir Lime2 access decision allowed without mandatory control - possible CVE-2023-22833 exploitation attempt</description>
    <group>access_control_bypass,</group>
  </rule>

  <rule id="100332" level="10" frequency="30" timeframe="300">
    <if_matched_sid>100331</if_matched_sid>
    <same_source_ip />
    <description>Multiple Palantir Lime2 access control bypass attempts from same source - CVE-2023-22833</description>
    <group>access_control_bypass,</group>
  </rule>

  <rule id="100333" level="8" frequency="20" timeframe="60">
    <decoded_as>json</decoded_as>
    <field name="user.privilege">LOW</field>
    <field name="resource.classification" type="pcre2">(?i)higher|top_secret|classified</field>
    <description>Low privilege Foundry user accessing high classification resources in burst - possible Lime2 bypass</description>
    <group>access_control_bypass,</group>
  </rule>
</group>

Mitigation & Patch Guidance

Patch Confidence Note: This advisory carries a patch confidence of 2/4 (CLAIMED), meaning that the CNA has stated a fix is available, but this has not been independently verified by other sources. Organizations should validate that the Limed2 version being deployed actually resolves the vulnerability before relying on the patch.

1) Immediate Mitigations (before patching):

• Review all user accounts in Foundry to identify and audit any users with administrative or project management roles; restrict to least privilege.
• Enable and review all audit logs related to access decision events, particularly those involving mandatory access control resources.
• If possible, temporarily restrict sensitive data access by adding additional discretionary access control restrictions to the most critical datasets as a defense-in-depth measure.
• Monitor for anomalous access patterns using the detection rules provided above.
• Consider whether any users have been recently provisioned with higher clearance levels; review their actions since provisioning.

2) Patch/Upgrade Steps:

• Contact Palantir support or your Palantir account team to request the patched Lime2 version.
• Verify the patched version number (likely 2.533.0 or later) directly with Palantir before deployment.
• Follow Palantir's standard deployment procedure for Foundry components, which typically involves staging, testing, and then deploying the upgraded Lime2 version.
• After patching, verify that the vulnerability is no longer present by attempting access workflows that previously triggered the bypass in a test environment (do not test on production).
• Confirm with Palantir representatives that the specific build deployed in your environment is not vulnerable to CVE-2023-22833.

3) Long-term Hardening:

• Review and tighten the mandatory access control policy configuration in Foundry to minimize the attack surface — restrict classification-level access to only those users who genuinely require it.
• Implement robust logging and alerting for all access control decisions in Foundry, with particular attention to anomalous access patterns.
• Conduct regular access control audits of the Foundry environment to ensure that discretionary and mandatory access control policies align with actual business requirements.
• Establish a relationship with Palantir's security advisory process to ensure timely notification of future vulnerabilities.
• Implement a least-privilege model across the entire Foundry deployment, including API keys and service accounts.

References

1. NVD Entry: https://nvd.nist.gov/vuln/detail/CVE-2023-22833 — Official NIST National Vulnerability Database entry containing the CNA description and metadata.
2. Palantir Security Advisories: https://www.palantir.com/security-advisories/ — Vendor security advisory channel; the official patch and advisory details should be available here.
3. Palantir Foundry Documentation — https://www.palantir.com/docs/foundry/ — Official documentation for Foundry access control architecture, providing context for understanding the impact of a MAC bypass.
4. CISA Known Exploited Vulnerabilities Catalog — https://www.cisa.gov/known-exploited-vulnerabilities-catalog — Confirms CVE-2023-22833 is NOT currently listed as exploited in the wild (as of the intelligence date).

Threat Actor Campaigns: No known threat actor campaigns have been publicly identified exploiting CVE-2023-22833, and there is no evidence of exploitation in the wild. The EPSS score of 0.00408 (33.7% percentile) indicates a relatively low probability of exploitation attempts in the near term. However, the access control bypass nature of this vulnerability makes it attractive for insider threat actors and espionage-focused groups, and the "certain circumstances" limitation in the description may be keeping exploit development at an early stage.