{"id":"CVE-2026-31237","kind":"vulnerability","aliases":["CVE-2026-31237","GHSA-wcr3-gm9f-f87q"],"related":[],"title":"Ludwig framework is vulnerable to insecure deserialization through its predict() method.","description":"The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization (CWE-502) through its predict() method. When a user provides a dataset file path to the predict() method, the framework automatically determines the file format. If the file is a pickle (.pkl) file, it is loaded using pandas.read_pickle() without any validation or security restrictions. This allows the deserialization of arbitrary Python objects via the unsafe pickle module. A remote attacker can exploit this by providing a maliciously crafted pickle file, leading to arbitrary code execution on the system running the Ludwig prediction.","summary":"Ludwig, a Python machine learning framework, loads a dataset passed to its predict() method with pandas.read_pickle() when the file ends in .pkl, with no validation. An attacker who supplies a malicious pickle file can run arbitrary Python code on the machine running the prediction.","severity":"critical","cvss":9.8,"epss":0.006,"exploited":false,"exploited_at":null,"products":[{"name":"ludwig","fixed":[],"affected":"<= 0.10.4","ecosystem":"pypi"}],"ecosystems":["pypi"],"tags":["cwe-502","ludwig","machine-learning","package","python"],"not_affected_if":["no-predict-on-external-files","trusted-input-only"],"how_to_check":"- Check the installed version with `pip show ludwig` and treat anything at or below `0.10.4` as affected.\n- Search your code and pipelines for calls to `predict()` that take a dataset path or file that could come from outside your control.","fix":"- Ludwig: unknown; the sources give no fixed version.","status":"open","published_at":"2026-05-12T18:16:52+00:00","first_seen_at":"2026-11-16T15:32:33+00:00","triaged_at":"2026-11-16T15:33:42+00:00","updated_at":"2026-11-16T15:37:59+00:00","category":"package","confidence":0.8,"triage":{"at":"2026-11-16T15:33:42Z","model":"claude-haiku-5-5","output":{"fix":"- Ludwig: unknown; the sources give no fixed version.","reach":"unknown","title":"Ludwig: predict() loads crafted pickle files, allowing arbitrary code execution","impact":"high","summary":"Ludwig, a Python machine learning framework, loads a dataset passed to its predict() method with pandas.read_pickle() when the file ends in .pkl, with no validation. An attacker who supplies a malicious pickle file can run arbitrary Python code on the machine running the prediction.","audience":"developers","category":"package","nicknames":[],"who_cares":["python","ludwig","machine-learning"],"confidence":0.8,"fix_status":"unknown","needs_auth":"unknown","worst_case":"Arbitrary code runs on the server doing the prediction.","analyst_note":"Confirm the fixed version in the GHSA advisory (https://github.com/advisories/GHSA-wcr3-gm9f-f87q) and whether a patched release exists, since the product data has no fixed version.","how_to_check":"- Check the installed version with `pip show ludwig` and treat anything at or below `0.10.4` as affected.\n- Search your code and pipelines for calls to `predict()` that take a dataset path or file that could come from outside your control.","plain_action":"","product_name":"","needs_analyst":true,"not_affected_if":["trusted-input-only","no-predict-on-external-files"],"who_is_affected":"Teams that run Ludwig's predict() on dataset files they do not fully trust, such as user uploads or files fetched from external sources, are exposed. Pipelines that only predict on datasets their own team produced and controls are lower risk. unknown whether the sources list other conditions.","user_interaction":"unknown","product_ecosystem":"","needs_special_setup":"no","severity_if_unknown":"critical"},"cost_usd":0.000776,"input_tokens":4698,"output_tokens":612,"severity_before":"critical"},"analyst_status":null,"analyst_note":null,"search":"'-2026':13A,21A '-31237':14A,22A '-502':25B,88C '0.10.4':81C 'allow':133C 'arbitrari':68B,137C,160C 'attack':59B,147C 'automat':107C 'code':70B,161C 'craft':155C 'cve':12A,20A 'cwe':24B,87C 'dataset':40B,98C 'deseri':7A,86C,135C 'determin':108C 'end':52B 'execut':162C 'exploit':149C 'f87q':19A 'file':51B,65B,99C,110C,114C,119C,157C 'format':111C 'framework':2A,37B,79C,106C 'ghsa':16A 'ghsa-wcr3-gm9f-f87q':15A 'gm9f':18A 'insecur':6A,85C 'lead':158C 'learn':29B,36B 'load':38B,122C 'ludwig':1A,23A,26B,32B,78C,168C 'machin':28B,35B,73B 'machine-learn':27B 'malici':63B,154C 'method':11A,45B,92C,104C 'modul':144C 'object':139C 'packag':30B 'pandas.read':47B,124C 'pass':41B 'path':100C 'pickl':48B,64B,117C,125C,143C,156C 'pkl':54B,118C 'predict':10A,44B,76B,91C,103C,169C 'provid':96C,152C 'python':31B,34B,69B,138C 'remot':146C 'restrict':131C 'run':67B,74B,166C 'secur':130C 'suppli':61B 'system':165C 'thru':80C 'unsaf':142C 'use':123C 'user':95C 'valid':57B,128C 'via':140C 'vulner':4A,83C 'wcr3':17A 'without':126C","nicknames":[],"mitigations":[],"who_affected":"Teams that run Ludwig's predict() on dataset files they do not fully trust, such as user uploads or files fetched from external sources, are exposed. Pipelines that only predict on datasets their own team produced and controls are lower risk. unknown whether the sources list other conditions.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"unknown","audience":"developers","plain_action":null,"pkeys":["ludwig"],"ptokens":["ludwig"],"labels":["critical","cwe502","ludwig","machinelearning","package","pypi","python"],"campaign_type":null,"search_a":"'-2026':13A,21A '-31237':14A,22A 'cve':12A,20A 'deseri':7A 'f87q':19A 'framework':2A 'ghsa':16A 'ghsa-wcr3-gm9f-f87q':15A 'gm9f':18A 'insecur':6A 'ludwig':1A,23A 'method':11A 'predict':10A 'vulner':4A 'wcr3':17A","search_b":"'-502':25B 'arbitrari':68B 'attack':59B 'code':70B 'cwe':24B 'dataset':40B 'end':52B 'file':51B,65B 'framework':37B 'learn':29B,36B 'load':38B 'ludwig':26B,32B 'machin':28B,35B,73B 'machine-learn':27B 'malici':63B 'method':45B 'packag':30B 'pandas.read':47B 'pass':41B 'pickl':48B,64B 'pkl':54B 'predict':44B,76B 'python':31B,34B,69B 'run':67B,74B 'suppli':61B 'valid':57B","search_c":"'-502':88C '0.10.4':81C 'allow':133C 'arbitrari':137C,160C 'attack':147C 'automat':107C 'code':161C 'craft':155C 'cwe':87C 'dataset':98C 'deseri':86C,135C 'determin':108C 'execut':162C 'exploit':149C 'file':99C,110C,114C,119C,157C 'format':111C 'framework':79C,106C 'insecur':85C 'lead':158C 'load':122C 'ludwig':78C,168C 'malici':154C 'method':92C,104C 'modul':144C 'object':139C 'pandas.read':124C 'path':100C 'pickl':117C,125C,143C,156C 'pkl':118C 'predict':91C,103C,169C 'provid':96C,152C 'python':138C 'remot':146C 'restrict':131C 'run':166C 'secur':130C 'system':165C 'thru':80C 'unsaf':142C 'use':123C 'user':95C 'valid':128C 'via':140C 'vulner':83C 'without':126C","cause_entry":null,"cause_path":null,"disclosures_checked_at":null,"first_public_at":"2026-05-12T18:16:52+00:00","kev":null,"cvss_vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H","cvss_by":"GitHub","poc_count":0,"sources":[{"source":"ghsa","external_id":"GHSA-wcr3-gm9f-f87q","url":"https://github.com/advisories/GHSA-wcr3-gm9f-f87q","severity":"critical","published_at":"2026-05-12T18:30:41+00:00","modified_at":"2026-11-16T15:31:38+00:00","first_seen_at":"2026-11-16T15:32:33+00:00","fetched_at":"2026-11-16T15:37:43+00:00","note":null},{"source":"nvd","external_id":"CVE-2026-31237","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-31237","severity":null,"published_at":"2026-05-12T18:16:52+00:00","modified_at":"2026-10-19T22:10:00+00:00","first_seen_at":"2026-11-16T15:37:59+00:00","fetched_at":"2026-11-16T15:37:59+00:00","note":null}],"documents":[],"disclosures":[],"exploit_sources":[{"type":"epss","score":0.006,"url":"https://api.first.org/data/v1/epss?cve=CVE-2026-31237"}],"notice":"Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai"}