NVIDIA Merlin Transformers4Rec para todas as plataformas contém uma vulnerabilidade em uma dependência do python, onde um atacante poderia causar um problema de injeção de código. Uma exploração bem-sucedida desta vulnerabilidade pode

The code injection vulnerability in NVIDIA Merlin Transformers4Rec presents a significant risk. An attacker could inject malicious code into the system, potentially gaining control over the affected environment. This could involve executing arbitrary commands with the privileges of the Merlin Transformers4Rec process, leading to data exfiltration, modification of recommendation models, or even complete system compromise. The impact is amplified if Merlin Transformers4Rec is deployed in a production environment handling sensitive user data or critical business processes. The ability to tamper with recommendation models could also lead to manipulation of user behavior and financial losses.

Organizations deploying NVIDIA Merlin Transformers4Rec for recommendation systems, particularly those handling sensitive user data or operating in environments with limited security controls, are at risk. Environments utilizing older, unpatched versions of Transformers4Rec are especially vulnerable.

• python / server:

import subprocess
result = subprocess.run(['pip', 'show', 'transformers4rec'], capture_output=True, text=True)
if 'Version' in result.stdout:
    version = result.stdout.split('Version: ')[1].split('\n')[0]
    if version.split('.')[0] == 'x.x' and int(version.split('.')[1]) < 1:
        print('Potential vulnerability: Transformers4Rec version is vulnerable.')
else:
    print('Transformers4Rec not found.')

1. 2025-08-13Disclosure

   disclosure

1. Reservado2025-01-14
2. Publicada2025-08-13
3. EPSS atualizado2026-03-23

The primary mitigation for CVE-2025-23298 is to upgrade to a version of NVIDIA Merlin Transformers4Rec that includes code commit b7eaea5. If an immediate upgrade is not feasible due to compatibility concerns or deployment constraints, consider isolating affected instances and limiting network access to minimize the attack surface. While a direct WAF rule is unlikely to be effective against code injection, carefully reviewing and restricting the inputs to the Python dependency could offer a limited layer of defense. Monitor system logs for unusual activity or unexpected code execution patterns.