CVE-2020-15208 High · 7.4 EPSS 0.9%

Data corruption in tensorflow-lite

tensorflow · tensorflow

Official description · CVE.org

In tensorflow-lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, when determining the common dimension size of two tensors, TFLite uses a `DCHECK` which is no-op outside of debug compilation modes. Since the function always returns the dimension of the first tensor, malicious attackers can craft cases where this is larger than that of the second tensor. In turn, this would result in reads/writes outside of bounds since the interpreter will wrongly assume that there is enough data in both tensors. The issue is patched in commit 8ee24e7949a203d234489f9da2c5bf45a7d5157d, and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.

Breakdown

17.1 / 100

Exploited: Not confirmed
Reports: 0
Tactics: 3/15

How an attack unfolds

Based on mappings. Select a step to see the evidence.

What they do
- Hijack how programs load codeT1574 · Hijack Execution FlowSource data8 general detection rules
- Crash the applicationT1499.004 · Application or System ExploitationSource data3 general detection rules

Disclosed on 25 Sep 2020. No public exploit code or confirmed exploitation is recorded.

techniques

2 , grouped by in kill-chain order 2

ExecutionTA0002

Hijack Execution Flow T1574 also Stealth

MITRE CTID mapping of the CVE · primary impact

ImpactTA0040

Application or System Exploitation T1499.004

MITRE CTID mapping of the CVE · primary impact

Detection

0 rules for this CVE · 11 general rules for its techniques

Rules for this CVE

0 that reference CVE-2020-15208 directly.

CVE-specific rules are not available in this build. Search SigmaHQ ↗

General rules for this CVE’s techniques

These detect the technique in general, not this specific vulnerability.

Hijack Execution FlowT15748 general rules

Exploiting SetupComplete.cmd CVE-2019-1378hightest
Potential PrintNightmare Exploitation Attempthightest
Suspicious Printer Driver Empty Manufacturerhightest
DLL Execution Via Register-cimprovider.exemediumtest
Potential Initial Access via DLL Search Order Hijackingmediumtest
Potential Registry Persistence Attempt Via DbgManagedDebuggermediumtest
Regsvr32 DLL Execution With Uncommon Extensionmediumtest
Windows Spooler Service Suspicious Binary Loadinformationaltest

Application or System ExploitationT1499.0043 general rules

Sigma rules © SigmaHQ, Detection Rule License 1.1; linked, not reproduced.

Am I affected? What do I upgrade to?

From GitHub Security Advisories and OSV

3 packages in 1 ecosystem. Main package: tensorflow.

PyPItensorflowGHSA-mxjj-953w-2c2v PYSEC-2020-131

Upgrade to 2.3.1 or later

Fixes this CVE and 2 other known vulnerabilities in this package (CVE-2020-15199, CVE-2020-15211).

pippip install "tensorflow>=2.3.1"

The first fix for CVE-2020-15208.

Affected versions, by release line

If you run	Upgrade to
2.3.0 up to 2.3.1	2.3.1
2.2.0 up to 2.2.1 Still affected on this line by CVE-2020-15199 (fixed in 2.3.1, a newer line).	2.2.1
2.1.0 up to 2.1.2 Still affected on this line by CVE-2020-15199 (fixed in 2.3.1, a newer line).	2.1.2
2.0.0 up to 2.0.3 Still affected on this line by CVE-2020-15199 (fixed in 2.3.1, a newer line).	2.0.3
All versions before 1.15.4 Still affected on this line by CVE-2020-15199 (fixed in 2.3.1, a newer line).	1.15.4

“Up to” means the second version is the fix and is not affected.

Other packages that bundle this code (2)

Forks, repackagings and products that include the affected code. Each has its own fixed version.

PyPItensorflow-cpu
Affected: 2.3.0 up to 2.3.1; 2.2.0 up to 2.2.1; 2.1.0 up to 2.1.2; 2.0.0 up to 2.0.3; All versions before 1.15.4Upgrade to 2.3.1 or later (older lines: 2.2.1, 2.1.2, 2.0.3, 1.15.4)
GHSA-mxjj-953w-2c2v+1
PyPItensorflow-gpu
Affected: 2.3.0 up to 2.3.1; 2.2.0 up to 2.2.1; 2.1.0 up to 2.1.2; 2.0.0 up to 2.0.3; All versions before 1.15.4Upgrade to 2.3.1 or later (older lines: 2.2.1, 2.1.2, 2.0.3, 1.15.4)
GHSA-mxjj-953w-2c2v+1

Check your lockfile against all CVEs

What security researchers reported

Not named in any threat report yet

References

4 links, grouped by NVD tag

Exploitation evidence

Not in KEV. CISA has not confirmed exploitation.

ENISA

Not marked exploited. EUVD-2020-0204

No public exploit code found.

, 30 Oct 2026

0.9%Top 40%

Chance of exploitation in the next 30 days. of scored CVEs (riskier than 59.8%).

3.1

7.4High · 7.4

CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:N

Priority factorsFormula

In a KEV catalognot listed 0.0/35

Recent KEV listingnot listed 0.0/10

Known ransomware usenot known 0.0/10

EPSS percentilepercentile 0.5977 11.9/20

Threat reports0 reports 0.0/10

Public exploit codenone found 0.0/5

Mentioned in OTX pulses0 pulses 0.0/3

CVSS base scorebase score 7.4 5.2/7

End-of-life productsupported or unknown 0.0

Total17.1

Record

Published: 25 Sep 2020
Modified: 4 Aug 2024
: CWE-125 Out-of-bounds ReadCWE-787 Out-of-bounds Write
Elsewhere: NVD CVE.org JSON

How an attack unfolds

1What they do

Exploitation timelineExploitation timelineDated milestones for a CVE: when it was disclosed, when exploit code became public, when exploitation was confirmed, and when threat reports named it.

ATT&CKMITRE ATT&CKA public catalogue of the techniques attackers use, maintained by MITRE. Patchfirst uses it to describe what attackers do with a vulnerability. techniques