CVE-2020-15194: TensorFlow: DoS via SparseFillEmptyRowsGrad assertion
MEDIUMPoC AVAILABLEA medium-severity denial-of-service vulnerability in TensorFlow's sparse gradient operation allows unauthenticated remote attackers to crash serving infrastructure by sending malformed tensor inputs. Any public-facing TensorFlow Serving endpoint that processes sparse tensor operations is at risk of availability disruption. Patch immediately to TF 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1 and restrict direct access to inference endpoints where possible.
What is the risk?
Medium risk with elevated operational impact for production ML serving environments. The network-accessible attack vector with no authentication required makes this trivially exploitable, though impact is limited to availability with no data exfiltration or code execution potential. Risk escalates significantly for organizations running TensorFlow Serving in production without input validation layers or network segmentation — a single crafted request is sufficient to crash the serving process and cause downtime.
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| TensorFlow | pip | — | No patch |
| leap | — | — | No patch |
How severe is it?
What is the attack surface?
What should I do?
5 stepsPatch: Upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1 — fix is commit 390611e0d45c5793c7066110af37c8514e6a6c54.
Network controls: Place an API gateway or reverse proxy in front of TF Serving; reject malformed or unexpected tensor shape payloads before they reach the TF runtime.
Input validation: Implement server-side shape validation for tensor inputs prior to gradient operations.
Detection: Monitor TF Serving process crash/restart rates and assertion failure logs as indicators of exploitation attempts.
Isolation: Run TF Serving in containers with automatic restart policies to reduce downtime impact if exploited before patching.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2020-15194?
A medium-severity denial-of-service vulnerability in TensorFlow's sparse gradient operation allows unauthenticated remote attackers to crash serving infrastructure by sending malformed tensor inputs. Any public-facing TensorFlow Serving endpoint that processes sparse tensor operations is at risk of availability disruption. Patch immediately to TF 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1 and restrict direct access to inference endpoints where possible.
Is CVE-2020-15194 actively exploited?
Proof-of-concept exploit code is publicly available for CVE-2020-15194, increasing the risk of exploitation.
How to fix CVE-2020-15194?
1. Patch: Upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1 — fix is commit 390611e0d45c5793c7066110af37c8514e6a6c54. 2. Network controls: Place an API gateway or reverse proxy in front of TF Serving; reject malformed or unexpected tensor shape payloads before they reach the TF runtime. 3. Input validation: Implement server-side shape validation for tensor inputs prior to gradient operations. 4. Detection: Monitor TF Serving process crash/restart rates and assertion failure logs as indicators of exploitation attempts. 5. Isolation: Run TF Serving in containers with automatic restart policies to reduce downtime impact if exploited before patching.
What systems are affected by CVE-2020-15194?
This vulnerability affects the following AI/ML architecture patterns: model serving, training pipelines.
What is the CVSS score for CVE-2020-15194?
CVE-2020-15194 has a CVSS v3.1 base score of 5.3 (MEDIUM). The EPSS exploitation probability is 1.03%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0029Denial of AI ServiceAML.T0043Craft Adversarial DataAML.T0049Exploit Public-Facing ApplicationCompliance Controls Affected
What are the technical details?
Original Advisory
In Tensorflow before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, the `SparseFillEmptyRowsGrad` implementation has incomplete validation of the shapes of its arguments. Although `reverse_index_map_t` and `grad_values_t` are accessed in a similar pattern, only `reverse_index_map_t` is validated to be of proper shape. Hence, malicious users can pass a bad `grad_values_t` to trigger an assertion failure in `vec`, causing denial of service in serving installations. The issue is patched in commit 390611e0d45c5793c7066110af37c8514e6a6c54, and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1."
Exploitation Scenario
An adversary identifies a public-facing TensorFlow Serving API — common in federated learning services or model retraining pipelines. Without authentication, they craft an inference request where grad_values_t carries a shape that passes top-level checks but triggers a shape mismatch assertion inside the vec() call of SparseFillEmptyRowsGrad. The serving process crashes immediately. With no rate limiting in place, the adversary scripts repeated requests to maintain a persistent DoS condition, keeping the ML inference service unavailable and potentially triggering costly auto-scaling loops or SLA breaches.
Weaknesses (CWE)
CWE-617 — Reachable Assertion: The product contains an assert() or similar statement that can be triggered by an attacker, which leads to an application exit or other behavior that is more severe than necessary.
- [Implementation] If the assertion statement is required (e.g., for security or reliability), change the invoking code to avoid the conditions that trigger the assertion.
- [Implementation] Consider changing the assertion to some other less "fatal" error. This might not be easy to implement, since other code may be needed to handle the error.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:LReferences
- lists.opensuse.org/opensuse-security-announce/2020-10/msg00065.htmlMailing List3rd Party
- github.com/tensorflow/tensorflow/commit/390611e0d45c5793c7066110af37c8514e6a6c54Patch3rd Party
- github.com/tensorflow/tensorflow/releases/tag/v2.3.1Release3rd Party
- github.com/tensorflow/tensorflow/security/advisories/GHSA-9mqp-7v2h-2382ExploitPatch3rd Party
Timeline
Related Vulnerabilities
CVE-2026-189489.9Feast: insecure UDF deserialization enables RCE
Same package: tensorflowCVE-2020-151969.9TensorFlow: heap OOB read in sparse/ragged count ops
Same package: tensorflowCVE-2020-152059.8TensorFlow: heap overflow in StringNGrams, ASLR bypass
Same package: tensorflowCVE-2019-167789.8TensorFlow: heap overflow in UnsortedSegmentSum op
Same package: tensorflowCVE-2020-152089.8TFLite: OOB read/write via tensor dimension mismatch
Same package: tensorflow