CVE-2025-71002 is a vulnerability identified in the flow.column_stack component of OneFlow version 0.9.0, a machine learning framework. The flaw is a floating-point exception (FPE) triggered by specially crafted inputs. This exception leads to a denial of service (DoS) condition, causing the affected process or service to crash or become unresponsive. The vulnerability is remotely exploitable over the network without requiring any privileges, but it does require user interaction, such as processing maliciously crafted data inputs. The CVSS v3.1 base score is 6.5, indicating medium severity, with the vector AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H, meaning no impact on confidentiality or integrity but a high impact on availability. The underlying weaknesses correspond to CWE-369 (Divide by Zero), CWE-704 (Incorrect Type Conversion or Cast), and CWE-681 (Incorrect Conversion between Numeric Types). No patches or fixes have been released at the time of publication, and no known exploits have been observed in the wild. The vulnerability could disrupt services relying on OneFlow for data stacking operations, potentially affecting AI workflows and data pipelines.

For European organizations, the primary impact is service disruption due to denial of service attacks targeting OneFlow deployments. This can affect availability of AI and data processing services, potentially delaying critical business operations, research, or analytics. Organizations relying on OneFlow in production environments, especially in sectors like finance, healthcare, and manufacturing where AI workloads are common, may face operational downtime. While confidentiality and integrity are not directly impacted, the loss of availability could indirectly affect business continuity and trust. The absence of patches increases risk exposure until mitigations are applied. Additionally, if attackers leverage this vulnerability in combination with other threats, it could facilitate more complex attack chains. The impact is more pronounced in organizations with externally facing AI services or those processing untrusted inputs without adequate filtering.

Given the lack of an official patch, European organizations should implement strict input validation and sanitization to prevent malformed data from reaching the flow.column_stack component. Network-level protections such as firewalls and intrusion detection systems should be configured to monitor and block suspicious traffic patterns targeting OneFlow services. Restrict access to OneFlow instances to trusted users and networks, minimizing exposure to untrusted inputs. Employ application-layer gateways or proxies to inspect and filter data before processing. Monitor logs and system behavior for signs of crashes or anomalies indicative of exploitation attempts. Consider isolating OneFlow workloads in containerized or sandboxed environments to limit impact of potential crashes. Engage with OneFlow maintainers and subscribe to security advisories for timely patch releases. Finally, develop incident response plans to quickly address DoS events affecting AI services.