NVIDIA

A critical vulnerability chain in NVIDIA's Triton Inference Server allows unauthenticated attackers to achieve complete remote code execution (RCE) and gain full control over AI servers. The attack exploits the server's Python backend through shared memory manipulation, leading to potential theft of proprietary AI models, exposure of sensitive data, and manipulation of AI responses. The vulnerability chain, identified as CVE-2025-23319, CVE-2025-23320, and CVE-2025-23334, poses significant risks to organizations using Triton for AI/ML operations, threatening intellectual property and operational security.

Source: https://cybersecuritynews.com/nvidia-triton-vulnerability/

TPRM report: https://www.rankiteo.com/company/nvidia

"id": "nvi251080925",
"linkid": "nvidia",
"type": "Vulnerability",
"date": "6/2025",
"severity": "100",
"impact": "5",
"explanation": "Attack threatening the organization's existence"

{'affected_entities': [{'industry': 'Technology',
                        'name': 'NVIDIA',
                        'type': 'Corporation'}],
 'attack_vector': 'Remote Code Execution (RCE) via shared memory manipulation',
 'data_breach': {'sensitivity_of_data': 'High',
                 'type_of_data_compromised': 'Proprietary AI models, sensitive '
                                             'data'},
 'date_publicly_disclosed': '2025-08-04',
 'description': 'A critical vulnerability chain in NVIDIA’s Triton Inference '
                'Server that allows unauthenticated attackers to achieve '
                'complete remote code execution (RCE) and gain full control '
                'over AI servers. The vulnerability chain, identified as '
                'CVE-2025-23319, CVE-2025-23320, and CVE-2025-23334, exploits '
                'the server’s Python backend through a sophisticated '
                'three-step attack process involving shared memory '
                'manipulation.',
 'impact': {'data_compromised': 'Proprietary AI models, sensitive data',
            'operational_impact': 'Theft of proprietary AI models, exposure of '
                                  'sensitive data, manipulation of AI model '
                                  'responses, network pivot points',
            'systems_affected': 'NVIDIA Triton Inference Server'},
 'initial_access_broker': {'entry_point': 'Shared memory manipulation via '
                                          '/dev/shm/',
                           'high_value_targets': 'AI models, sensitive data'},
 'investigation_status': 'Patches released',
 'post_incident_analysis': {'corrective_actions': 'Patches released in Triton '
                                                  'Inference Server version '
                                                  '25.07',
                            'root_causes': 'Lack of proper validation in '
                                           'shared memory API, information '
                                           'disclosure via error messages'},
 'recommendations': 'Update to Triton Inference Server version 25.07 '
                    'immediately',
 'references': [{'source': 'Wiz Research'}],
 'response': {'containment_measures': 'Patches released in Triton Inference '
                                      'Server version 25.07',
              'remediation_measures': 'Update to Triton Inference Server '
                                      'version 25.07',
              'third_party_assistance': 'Wiz Research'},
 'title': 'Critical Vulnerability Chain in NVIDIA Triton Inference Server',
 'type': 'Vulnerability Exploitation',
 'vulnerability_exploited': ['CVE-2025-23319',
                             'CVE-2025-23320',
                             'CVE-2025-23334']}