Text Embeddings Inference EUVD-2026-97008

Exploitation requires the attacker to already have read access to one of the data sinks where the router prints its configuration: the router startup logs, container stdout/output, or the OTLP telemetry stream - and the deployment must have an api_key configured (otherwise there is no protected endpoint and no secret to leak). It is specifically the router's Args struct lacking a redact attribute for the api_key field that causes the key to be logged in cleartext. Limiting factors: this is not remotely exploitable over the network (AV:L) and requires existing privileged/log-reading access (PR:L); the impact is confined to recovering one Bearer token that grants access only to the embedding and rerank endpoints, with no direct integrity or availability effect.

This is a genuine but modest-severity information-disclosure flaw (CWE-532), not a high-priority emergency. The only severity signal provided is the vendor/CVE CVSS 4.0 vector (CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:N, base 4.8 Medium), which is internally consistent with the description: AV:L and PR:L reflect that an attacker must already have access to router logs, container stdout, or OTLP telemetry - not a remote network path - and VC:L reflects limited confidentiality impact (leak of one Bearer token) with no integrity or availability impact. No EPSS score, CISA KEV listing, POC reference, or SSVC decision point was supplied, so exploitation-probability and automatability signals are simply absent rather than conflicting. There is no public exploit identified at time of analysis. The practical risk is bounded: an attacker who can already read the service's logs/telemetry is usually already inside a trust boundary, and the recovered key only unlocks the embedding and rerank endpoints (not arbitrary code or data exfiltration). Treat as a routine patch/log-hygiene item, prioritized higher only in multi-tenant or shared-logging environments where log access is broader than endpoint access.

In a shared-observability deployment, a developer or operator with read access to the Text Embeddings Inference router's startup logs, container stdout, or OTLP telemetry pipeline reads the api_key printed in cleartext at startup. They reuse that Bearer token to call the otherwise-protected embedding and rerank endpoints, bypassing the intended authentication. No public exploit code is required - recovery is a simple log/telemetry read.