{"id":"CVE-2026-31238","kind":"vulnerability","aliases":["CVE-2026-31238","GHSA-xp5q-5q7g-q26r"],"related":[],"title":"Ludwig framework is vulnerable to insecure deserialization in its model serving component","description":"The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization (CWE-502) in its model serving component. When starting a model server with the ludwig serve command, the framework loads model weight files using torch.load() without enabling the security-restrictive weights_only=True parameter. This default behavior allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing a maliciously crafted PyTorch model file, leading to arbitrary code execution on the system hosting the Ludwig model server.","summary":"Ludwig through 0.10.4 loads model weight files with torch.load() without weights_only=True when running the ludwig serve command. A crafted PyTorch model file can run arbitrary Python code on the server hosting the model.","severity":"critical","cvss":9.8,"epss":0.00497,"exploited":false,"exploited_at":null,"products":[{"name":"ludwig","fixed":[],"affected":"<= 0.10.4","ecosystem":"pypi"}],"ecosystems":["pypi"],"tags":["cwe-502","ludwig","ml-serving","package","python"],"not_affected_if":["no-ludwig-serve","trusted-model-files-only"],"how_to_check":"- Check the installed version with `pip show ludwig` and compare it to 0.10.4.\n- Look for servers started with `ludwig serve` that load model files from paths or sources outside your control.","fix":"- ludwig: fixed version unknown; the sources do not state one.","status":"open","published_at":"2026-05-12T18:16:52+00:00","first_seen_at":"2026-11-16T15:22:27+00:00","triaged_at":"2026-11-16T15:23:24+00:00","updated_at":"2026-11-16T15:37:59+00:00","category":"package","confidence":0.7,"triage":{"at":"2026-11-16T15:23:24Z","model":"claude-haiku-5-5","output":{"fix":"- ludwig: fixed version unknown; the sources do not state one.","reach":"unknown","title":"Ludwig: model serving loads untrusted weights with pickle, enabling code execution","impact":"high","summary":"Ludwig through 0.10.4 loads model weight files with torch.load() without weights_only=True when running the ludwig serve command. A crafted PyTorch model file can run arbitrary Python code on the server hosting the model.","audience":"developers","category":"package","nicknames":[],"who_cares":["python","ludwig","ml-serving"],"confidence":0.7,"fix_status":"unknown","needs_auth":"unknown","worst_case":"Arbitrary code runs on the server hosting the model","analyst_note":"Find the fixed version in the GHSA-xp5q-5q7g-q26r advisory at https://github.com/advisories/GHSA-xp5q-5q7g-q26r and confirm whether the reach is network or local.","how_to_check":"- Check the installed version with `pip show ludwig` and compare it to 0.10.4.\n- Look for servers started with `ludwig serve` that load model files from paths or sources outside your control.","plain_action":"","product_name":"","needs_analyst":true,"not_affected_if":["no-ludwig-serve","trusted-model-files-only"],"who_is_affected":"Teams that run the ludwig serve command to host Ludwig models and load weight files that could come from an untrusted source. Applications that only train models locally or never serve weight files from outside the team are less exposed. The sources do not say whether the server itself must be reachable from the internet.","user_interaction":"unknown","product_ecosystem":"","needs_special_setup":"no","severity_if_unknown":"critical"},"cost_usd":0.000757,"input_tokens":4678,"output_tokens":578,"severity_before":"critical"},"analyst_status":null,"analyst_note":null,"search":"'-2026':14A,22A '-31238':15A,23A '-502':26B,79C '0.10.4':35B,72C '5q7g':19A 'allow':116C 'arbitrari':59B,120C,142C 'attack':128C 'behavior':115C 'code':61B,143C 'command':51B,94C 'compon':12A,84C 'craft':53B,136C 'cve':13A,21A 'cwe':25B,78C 'default':114C 'deseri':7A,77C,118C 'enabl':104C 'execut':144C 'exploit':130C 'file':39B,56B,100C,139C 'framework':2A,70C,96C 'ghsa':17A 'ghsa-xp5q-5q7g-q26r':16A 'host':65B,148C 'insecur':6A,76C 'lead':140C 'load':36B,97C 'ludwig':1A,24A,27B,33B,49B,69C,92C,150C 'malici':135C 'ml':29B 'ml-serv':28B 'model':10A,37B,55B,67B,82C,88C,98C,138C,151C 'modul':126C 'object':122C 'packag':31B 'paramet':112C 'pickl':125C 'provid':133C 'python':32B,60B,121C 'pytorch':54B,137C 'q26r':20A 'restrict':108C 'run':47B,58B 'secur':107C 'security-restrict':106C 'serv':11A,30B,50B,83C,93C 'server':64B,89C,152C 'start':86C 'system':147C 'thru':71C 'torch.load':41B,102C 'true':45B,111C 'use':101C 'via':123C 'vulner':4A,74C 'weight':38B,43B,99C,109C 'without':42B,103C 'xp5q':18A","nicknames":[],"mitigations":[],"who_affected":"Teams that run the ludwig serve command to host Ludwig models and load weight files that could come from an untrusted source. Applications that only train models locally or never serve weight files from outside the team are less exposed. The sources do not say whether the server itself must be reachable from the internet.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"unknown","audience":"developers","plain_action":null,"pkeys":["ludwig"],"ptokens":["ludwig"],"labels":["critical","cwe502","ludwig","mlserving","package","pypi","python"],"campaign_type":null,"search_a":"'-2026':14A,22A '-31238':15A,23A '5q7g':19A 'compon':12A 'cve':13A,21A 'deseri':7A 'framework':2A 'ghsa':17A 'ghsa-xp5q-5q7g-q26r':16A 'insecur':6A 'ludwig':1A,24A 'model':10A 'q26r':20A 'serv':11A 'vulner':4A 'xp5q':18A","search_b":"'-502':26B '0.10.4':35B 'arbitrari':59B 'code':61B 'command':51B 'craft':53B 'cwe':25B 'file':39B,56B 'host':65B 'load':36B 'ludwig':27B,33B,49B 'ml':29B 'ml-serv':28B 'model':37B,55B,67B 'packag':31B 'python':32B,60B 'pytorch':54B 'run':47B,58B 'serv':30B,50B 'server':64B 'torch.load':41B 'true':45B 'weight':38B,43B 'without':42B","search_c":"'-502':79C '0.10.4':72C 'allow':116C 'arbitrari':120C,142C 'attack':128C 'behavior':115C 'code':143C 'command':94C 'compon':84C 'craft':136C 'cwe':78C 'default':114C 'deseri':77C,118C 'enabl':104C 'execut':144C 'exploit':130C 'file':100C,139C 'framework':70C,96C 'host':148C 'insecur':76C 'lead':140C 'load':97C 'ludwig':69C,92C,150C 'malici':135C 'model':82C,88C,98C,138C,151C 'modul':126C 'object':122C 'paramet':112C 'pickl':125C 'provid':133C 'python':121C 'pytorch':137C 'restrict':108C 'secur':107C 'security-restrict':106C 'serv':83C,93C 'server':89C,152C 'start':86C 'system':147C 'thru':71C 'torch.load':102C 'true':111C 'use':101C 'via':123C 'vulner':74C 'weight':99C,109C 'without':103C","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-xp5q-5q7g-q26r","url":"https://github.com/advisories/GHSA-xp5q-5q7g-q26r","severity":"critical","published_at":"2026-05-12T18:30:41+00:00","modified_at":"2026-11-16T15:20:40+00:00","first_seen_at":"2026-11-16T15:22:27+00:00","fetched_at":"2026-11-16T15:27:33+00:00","note":null},{"source":"nvd","external_id":"CVE-2026-31238","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-31238","severity":null,"published_at":"2026-05-12T18:16:52+00:00","modified_at":"2026-06-17T10:33:29+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":0.00497,"url":"https://api.first.org/data/v1/epss?cve=CVE-2026-31238"}],"notice":"Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai"}