{"id":"CVE-2020-15211","aliases":["BIT-tensorflow-2020-15211","EUVD-2020-0207","GHSA-cvpc-8phh-8f45","PYSEC-2020-134","PYSEC-2020-291","PYSEC-2020-326","openSUSE-SU-2020:1766-1"],"status":"published","title":"Out of bounds access in tensorflow-lite","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","severity":{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium","assessor":"cna","from":"cve"},"epss":{"day":"2026-11-09","score":0.00917,"percentile":0.60471,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":false,"fix_available":true},"cwes":["CWE-125","CWE-787"],"published":"2020-09-25T18:28:49Z","modified":"2026-10-20T14:10:35Z","fixes":[{"ecosystem":"PyPI","package":"tensorflow","fixed_in":["1.15.4","2.0.3","2.1.2","2.2.1","2.3.1"],"sources":["ghsa","osv"]},{"ecosystem":"PyPI","package":"tensorflow-cpu","fixed_in":["1.15.4","2.0.3","2.1.2","2.2.1","2.3.1"],"sources":["ghsa","osv"]},{"ecosystem":"PyPI","package":"tensorflow-gpu","fixed_in":["1.15.4","2.0.3","2.1.2","2.2.1","2.3.1"],"sources":["ghsa","osv"]},{"ecosystem":"Bitnami","package":"tensorflow","fixed_in":["1.15.4","2.0.3","2.1.2","2.2.1","2.3.1"],"sources":["osv"]}],"affected":[{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"last_affected":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"last_affected":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"last_affected":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"last_affected":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"last_affected":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"ghsa","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"last_affected":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"osv","ecosystem":"Bitnami","package":"tensorflow","ranges":[{"type":"SEMVER","events":[{"introduced":"0"},{"fixed":"1.15.4"},{"introduced":"2.0.0"},{"fixed":"2.0.3"},{"introduced":"2.1.0"},{"fixed":"2.1.2"},{"introduced":"2.2.0"},{"fixed":"2.2.1"},{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}],"versions":["0.12.0","0.12.1","1.0.0","1.0.1","1.1.0","1.10.0","1.10.1","1.11.0","1.12.0","1.12.2","1.12.3","1.13.1","1.13.2","1.14.0","1.15.0","1.15.2","1.15.3","1.2.0","1.2.1","1.3.0","1.4.0","1.4.1","1.5.0","1.5.1","1.6.0","1.7.0","1.7.1","1.8.0","1.9.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}],"versions":["2.0.0","2.0.1","2.0.2"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}],"versions":["2.1.0","2.1.1"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}],"versions":["1.15.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}],"versions":["2.1.0","2.1.1"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"}]}],"versions":["0.12.0","0.12.1","1.0.0","1.0.1","1.1.0","1.10.0","1.10.1","1.11.0","1.12.0","1.12.2","1.12.3","1.13.1","1.13.2","1.14.0","1.15.0","1.15.2","1.15.3","1.2.0","1.2.1","1.3.0","1.4.0","1.4.1","1.5.0","1.5.1","1.6.0","1.7.0","1.7.1","1.8.0","1.9.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.0.0"},{"fixed":"2.0.3"}]}],"versions":["2.0.0","2.0.1","2.0.2"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.1.0"},{"fixed":"2.1.2"}]}],"versions":["2.1.0","2.1.1"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.2.0"},{"fixed":"2.2.1"}]}],"versions":["2.2.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["2.3.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow","ranges":[{"type":"GIT","events":[{"introduced":"0"},{"fixed":"e11f55585f614645b360563072ffeb5c3eeff162"},{"fixed":"cd31fd0ce0449a9e0f83dcad08d6ed7f1d6bef3f"},{"fixed":"46d5b0852528ddfd614ded79bccc75589f801bd9"},{"fixed":"00302787b788c5ff04cb6f62aed5a74d936e86c0"},{"fixed":"fff2c8326280c07733828f990548979bdc893859"},{"fixed":"1970c2158b1ffa416d159d03c3370b9a462aee35"}]},{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"},{"introduced":"2.0.0"},{"fixed":"2.0.3"},{"introduced":"2.1.0"},{"fixed":"2.1.2"},{"introduced":"2.2.0"},{"fixed":"2.2.1"},{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["0.12.0rc0","0.12.0rc1","0.12.0","0.12.1","1.0.0","1.0.1","1.1.0rc0","1.1.0rc1","1.1.0rc2","1.1.0","1.2.0rc0","1.2.0rc1","1.2.0rc2","1.2.0","1.2.1","1.3.0rc0","1.3.0rc1","1.3.0rc2","1.3.0","1.4.0rc0","1.4.0rc1","1.4.0","1.4.1","1.5.0rc0","1.5.0rc1","1.5.0","1.5.1","1.6.0rc0","1.6.0rc1","1.6.0","1.7.0rc0","1.7.0rc1","1.7.0","1.7.1","1.8.0rc0","1.8.0rc1","1.8.0","1.9.0rc0","1.9.0rc1","1.9.0rc2","1.9.0","1.10.0rc0","1.10.0rc1","1.10.0","1.10.1","1.11.0rc0","1.11.0rc1","1.11.0rc2","1.11.0","1.12.0rc0","1.12.0rc1","1.12.0rc2","1.12.0","1.12.2","1.12.3","1.13.0rc0","1.13.0rc1","1.13.0rc2","1.13.1","1.13.2","1.14.0rc0","1.14.0rc1","1.14.0","1.15.0rc0","1.15.0rc1","1.15.0rc2","1.15.0rc3","1.15.0","1.15.2","1.15.3","2.0.0","2.0.1","2.0.2","2.1.0","2.1.1","2.2.0","2.3.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-cpu","ranges":[{"type":"GIT","events":[{"introduced":"0"},{"fixed":"e11f55585f614645b360563072ffeb5c3eeff162"},{"fixed":"cd31fd0ce0449a9e0f83dcad08d6ed7f1d6bef3f"},{"fixed":"46d5b0852528ddfd614ded79bccc75589f801bd9"},{"fixed":"00302787b788c5ff04cb6f62aed5a74d936e86c0"},{"fixed":"fff2c8326280c07733828f990548979bdc893859"},{"fixed":"1970c2158b1ffa416d159d03c3370b9a462aee35"}]},{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"},{"introduced":"2.0.0"},{"fixed":"2.0.3"},{"introduced":"2.1.0"},{"fixed":"2.1.2"},{"introduced":"2.2.0"},{"fixed":"2.2.1"},{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["1.15.0","2.1.0","2.1.1","2.2.0","2.3.0"]},{"source":"osv","ecosystem":"PyPI","package":"tensorflow-gpu","ranges":[{"type":"GIT","events":[{"introduced":"0"},{"fixed":"e11f55585f614645b360563072ffeb5c3eeff162"},{"fixed":"cd31fd0ce0449a9e0f83dcad08d6ed7f1d6bef3f"},{"fixed":"46d5b0852528ddfd614ded79bccc75589f801bd9"},{"fixed":"00302787b788c5ff04cb6f62aed5a74d936e86c0"},{"fixed":"fff2c8326280c07733828f990548979bdc893859"},{"fixed":"1970c2158b1ffa416d159d03c3370b9a462aee35"}]},{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"1.15.4"},{"introduced":"2.0.0"},{"fixed":"2.0.3"},{"introduced":"2.1.0"},{"fixed":"2.1.2"},{"introduced":"2.2.0"},{"fixed":"2.2.1"},{"introduced":"2.3.0"},{"fixed":"2.3.1"}]}],"versions":["0.12.0","0.12.1","1.0.0","1.0.1","1.1.0","1.10.0","1.10.1","1.11.0","1.12.0","1.12.2","1.12.3","1.13.1","1.13.2","1.14.0","1.15.0","1.15.2","1.15.3","1.2.0","1.2.1","1.3.0","1.4.0","1.4.1","1.5.0","1.5.1","1.6.0","1.7.0","1.7.1","1.8.0","1.9.0","2.0.0","2.0.1","2.0.2","2.1.0","2.1.1","2.2.0","2.3.0"]},{"source":"osv","ecosystem":"openSUSE:Leap 15.2","package":"tensorflow2","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"2.1.2-lp152.7.3.1"}]}],"bundled":16},{"source":"osv","ecosystem":"openSUSE:Leap 15.2","package":"tensorflow2-lite","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"2.1.2-lp152.7.3.1"}]}],"bundled":16},{"source":"osv","ecosystem":"openSUSE:Leap 15.2","package":"tensorflow2_2_1_2-gnu-hpc","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"2.1.2-lp152.7.3.1"}]}],"bundled":16},{"source":"osv","ecosystem":"openSUSE:Leap 15.2","package":"tensorflow2_2_1_2-gnu-openmpi2-hpc","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"2.1.2-lp152.7.3.1"}]}],"bundled":16}],"products":[{"vendor":"google","product":"tensorflow"},{"vendor":"opensuse","product":"leap"},{"vendor":"tensorflow","product":"tensorflow"}],"references":[{"url":"https://github.com/tensorflow/tensorflow/releases/tag/v2.3.1","tags":["x_refsource_MISC","x_transferred","Third Party Advisory","advisory","web"]},{"url":"https://github.com/tensorflow/tensorflow/security/advisories/GHSA-cvpc-8phh-8f45","tags":["x_refsource_CONFIRM","x_transferred","Exploit","Third Party Advisory","advisory","evidence","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/46d5b0852528ddfd614ded79bccc75589f801bd9","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/00302787b788c5ff04cb6f62aed5a74d936e86c0","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/e11f55585f614645b360563072ffeb5c3eeff162","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/cd31fd0ce0449a9e0f83dcad08d6ed7f1d6bef3f","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/1970c2158b1ffa416d159d03c3370b9a462aee35","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/fff2c8326280c07733828f990548979bdc893859","tags":["x_refsource_MISC","x_transferred","Patch","Third Party Advisory","advisory","fix","web"]},{"url":"http://lists.opensuse.org/opensuse-security-announce/2020-10/msg00065.html","tags":["vendor-advisory","x_refsource_SUSE","x_transferred","Mailing List","Third Party Advisory","advisory","web"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2020-15211","tags":["web","advisory"]},{"url":"https://github.com/tensorflow/tensorflow/commit/f911af101dc0ce0eec17a8740bec9b613ae4195e","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/e6b213cebb56f485bd400961a2ed109aeeac9d3c","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/e47eb1453f35666795a31e208c28922b08756c69","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/d8f8236c29744b8e3247c083fd21c9a87180505c","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/c22736982844d19af623ccd7d33e2d199493eee7","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/7e283f97d8c784d3eae5062d9de25d0f432ad239","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/42ed6ac86856956da65b5957a26fab130ff9471c","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/38cbad757b2e1c0d64b95e4582408fa66627a67c","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/1a8528bfb572884eb8137dab1bf649705c960c47","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/commit/0b5be2717a19ca7bf505369eb8bdd341405d263d","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/kernels/kernel_util.cc#L36","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/core/subgraph.cc#L566-L580","tags":["web"]},{"url":"https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/c/common.h#L82","tags":["web"]},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow/PYSEC-2020-134.yaml","tags":["web"]},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow-gpu/PYSEC-2020-326.yaml","tags":["web"]},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow-cpu/PYSEC-2020-291.yaml","tags":["web"]},{"url":"https://github.com/advisories/GHSA-cvpc-8phh-8f45"},{"url":"https://github.com/tensorflow/tensorflow","tags":["package"]},{"url":"https://lists.opensuse.org/archives/list/security-announce@lists.opensuse.org/thread/TNMEEN772D6LWSHNB64QFB5TB3CZZEF4/","tags":["advisory"]},{"url":"https://bugzilla.suse.com/1173314","tags":["report"]},{"url":"https://bugzilla.suse.com/1175099","tags":["report"]},{"url":"https://bugzilla.suse.com/1175789","tags":["report"]},{"url":"https://bugzilla.suse.com/1177022","tags":["report"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15190","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15191","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15192","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15193","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15194","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15195","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15202","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15203","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15204","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15205","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15206","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15207","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15208","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15209","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15210","tags":["web"]},{"url":"https://www.suse.com/security/cve/CVE-2020-15211","tags":["web"]}],"sources":{"cve":[{"id":"CVE-2020-15211","title":"Out of bounds access in tensorflow-lite","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T18:45:24Z","modified":"2024-08-04T13:08:22.936Z","cvss":[{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium","assessor":"cna"}],"cwes":["CWE-125","CWE-787"],"updated_at":"2026-11-02T14:35:50.228781Z"}],"euvd":[{"id":"EUVD-2020-0207","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T18:45:24Z","modified":"2024-08-04T13:08:22Z","cvss":[{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium"}],"updated_at":"2026-11-03T03:07:33.100327Z"}],"ghsa":[{"id":"GHSA-cvpc-8phh-8f45","title":"Out of bounds access in tensorflow-lite","description":"### Impact\nIn TensorFlow Lite, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor: https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/kernels/kernel_util.cc#L36\n\nHowever, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors:\nhttps://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/c/common.h#L82\n\nThis results in special casing during validation at model loading time: https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/core/subgraph.cc#L566-L580\n\nUnfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays.\n\nThis results in both read and write gadgets, albeit very limited in scope.\n\n### Patches\nWe have patched the issue in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83). We will release patch releases for all versions between 1.15 and 2.3.\n\nWe recommend users to upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.\n\n### Workarounds\nA potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.\n\n### For more information\nPlease consult [our security guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for more information regarding the security model and how to contact us with issues and questions.\n\n### Attribution\nThis vulnerability has been reported by members of the Aivul Team from Qihoo 360.","status":"published","published":"2020-09-25T18:28:49Z","modified":"2024-10-28T15:02:08Z","cvss":[{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N","level":"medium"},{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium"}],"cwes":["CWE-125","CWE-787"],"updated_at":"2026-11-02T14:38:14.769863Z"}],"nvd":[{"id":"CVE-2020-15211","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T19:15:16.4Z","modified":"2026-06-17T02:56:16.59Z","cvss":[{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium","assessor":"secondary"},{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium","assessor":"primary"},{"version":"2.0","score":5.8,"vector":"AV:N/AC:M/Au:N/C:P/I:P/A:N","level":"medium","assessor":"primary"}],"cwes":["CWE-125","CWE-787"],"updated_at":"2026-11-02T14:06:17.70091Z"}],"osv":[{"id":"CVE-2020-15211","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T19:15:16.4Z","modified":"2026-10-20T14:10:35.436662643Z","cvss":[{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium"}],"updated_at":"2026-11-02T15:44:14.828228Z"},{"id":"GHSA-cvpc-8phh-8f45","title":"Out of bounds access in tensorflow-lite","description":"### Impact\nIn TensorFlow Lite, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor: https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/kernels/kernel_util.cc#L36\n\nHowever, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors:\nhttps://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/c/common.h#L82\n\nThis results in special casing during validation at model loading time: https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/core/subgraph.cc#L566-L580\n\nUnfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays.\n\nThis results in both read and write gadgets, albeit very limited in scope.\n\n### Patches\nWe have patched the issue in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83). We will release patch releases for all versions between 1.15 and 2.3.\n\nWe recommend users to upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.\n\n### Workarounds\nA potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.\n\n### For more information\nPlease consult [our security guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for more information regarding the security model and how to contact us with issues and questions.\n\n### Attribution\nThis vulnerability has been reported by members of the Aivul Team from Qihoo 360.","status":"published","published":"2020-09-25T18:28:49Z","modified":"2026-07-08T06:27:51.855493584Z","cvss":[{"version":"3.1","score":4.8,"vector":"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N","level":"medium"},{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N","level":"medium"}],"cwes":["CWE-125","CWE-787"],"updated_at":"2026-11-02T15:46:45.639885Z"},{"id":"BIT-tensorflow-2020-15211","title":"Out of bounds access in tensorflow-lite","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2024-03-06T11:20:26.16Z","modified":"2026-09-08T08:48:19.887872633Z","updated_at":"2026-11-02T15:28:35.51001Z"},{"id":"openSUSE-SU-2020:1766-1","title":"Security update for tensorflow2","description":"This update for tensorflow2 fixes the following issues:\n\n- updated to 2.1.2 with following fixes (boo#1177022):\n  * Fixes an undefined behavior causing a segfault in tf.raw_ops.Switch (CVE-2020-15190)\n  * Fixes three vulnerabilities in conversion to DLPack format (CVE-2020-15191, CVE-2020-15192, CVE-2020-15193)\n  * Fixes two vulnerabilities in SparseFillEmptyRowsGrad (CVE-2020-15194, CVE-2020-15195)\n  * Fixes an integer truncation vulnerability in code using the work sharder API (CVE-2020-15202)\n  * Fixes a format string vulnerability in tf.strings.as_string (CVE-2020-15203)\n  * Fixes segfault raised by calling session-only ops in eager mode (CVE-2020-15204)\n  * Fixes data leak and potential ASLR violation from tf.raw_ops.StringNGrams (CVE-2020-15205)\n  * Fixes segfaults caused by incomplete SavedModel validation (CVE-2020-15206)\n  * Fixes a data corruption due to a bug in negative indexing support in TFLite (CVE-2020-15207)\n  * Fixes a data corruption due to dimension mismatch in TFLite (CVE-2020-15208)\n  * Fixes several vulnerabilities in TFLite saved model format (CVE-2020-15209, CVE-2020-15210, CVE-2020-15211)\n","status":"published","published":"2020-10-29T11:23:39Z","modified":"2026-02-04T03:45:35.673536Z","updated_at":"2026-11-02T16:58:31.517538Z"},{"id":"PYSEC-2020-134","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T19:15:00Z","modified":"2023-12-06T01:00:17.303391Z","updated_at":"2026-11-02T15:52:52.192131Z"},{"id":"PYSEC-2020-291","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T19:15:00Z","modified":"2023-12-06T01:00:17.303391Z","updated_at":"2026-11-02T15:52:52.682031Z"},{"id":"PYSEC-2020-326","description":"In TensorFlow Lite before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, saved models in the flatbuffer format use a double indexing scheme: a model has a set of subgraphs, each subgraph has a set of operators and each operator has a set of input/output tensors. The flatbuffer format uses indices for the tensors, indexing into an array of tensors that is owned by the subgraph. This results in a pattern of double array indexing when trying to get the data of each tensor. However, some operators can have some tensors be optional. To handle this scenario, the flatbuffer model uses a negative `-1` value as index for these tensors. This results in special casing during validation at model loading time. Unfortunately, this means that the `-1` index is a valid tensor index for any operator, including those that don't expect optional inputs and including for output tensors. Thus, this allows writing and reading from outside the bounds of heap allocated arrays, although only at a specific offset from the start of these arrays. This results in both read and write gadgets, albeit very limited in scope. The issue is patched in several commits (46d5b0852, 00302787b7, e11f5558, cd31fd0ce, 1970c21, and fff2c83), and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1. A potential workaround would be to add a custom `Verifier` to the model loading code to ensure that only operators which accept optional inputs use the `-1` special value and only for the tensors that they expect to be optional. Since this allow-list type approach is erro-prone, we advise upgrading to the patched code.","status":"published","published":"2020-09-25T19:15:00Z","modified":"2023-12-06T01:00:17.303391Z","updated_at":"2026-11-02T15:52:52.682031Z"}]}}