{"id":"CVE-2026-123653","kind":"vulnerability","aliases":["CVE-2026-123653"],"related":[],"title":"DeepSeek-Reasonix: prompt injection through document discovery lets attackers steer the model","description":"A security vulnerability has been detected in esengine DeepSeek-Reasonix up to 2.52.0. Affected is the function discoverDocs of the file internal/memory/doc.go. The manipulation leads to improper neutralization of input used for llm prompting. The attack can be initiated remotely. The exploit has been disclosed publicly and may be used. Upgrading to version 2.53.0 is able to address this issue. The identifier of the patch is 4ca33493a673c2788aa8598fbca2b2f47a69e1b8. It is suggested to upgrade the affected component.","summary":"DeepSeek-Reasonix, an AI coding tool, fails to neutralize input in its discoverDocs function, so crafted content can inject instructions into the LLM prompt. The sources say an attack can be started remotely and that a public exploit has been disclosed.","severity":"high","cvss":7.5,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["ai-coding-tools","ai-tooling","deepseek-reasonix","llm-prompting"],"not_affected_if":["version-2.53.0-or-later"],"how_to_check":"- Check the installed version of DeepSeek-Reasonix; versions up to `2.52.0` are affected.\n- Look at whether the tool reads project documents through `internal/memory/doc.go` (`discoverDocs`) in your build.","fix":"- DeepSeek-Reasonix: upgrade to `2.53.0`, per the sources.","status":"open","published_at":"2026-11-06T14:17:10+00:00","first_seen_at":"2026-11-06T14:42:11+00:00","triaged_at":"2026-11-06T14:43:37+00:00","updated_at":"2026-11-06T14:43:37+00:00","category":"ai-tooling","confidence":0.6,"triage":{"at":"2026-11-06T14:43:37Z","model":"claude-haiku-5-5","output":{"fix":"- DeepSeek-Reasonix: upgrade to `2.53.0`, per the sources.","reach":"internet","title":"DeepSeek-Reasonix: prompt injection through document discovery lets attackers steer the model","impact":"medium","summary":"DeepSeek-Reasonix, an AI coding tool, fails to neutralize input in its discoverDocs function, so crafted content can inject instructions into the LLM prompt. The sources say an attack can be started remotely and that a public exploit has been disclosed.","audience":"developers","category":"ai-tooling","nicknames":[],"who_cares":["deepseek-reasonix","ai-coding-tools","llm-prompting"],"confidence":0.6,"fix_status":"available","needs_auth":"unknown","worst_case":"injected prompts steer the model's actions in the coding agent","analyst_note":"Check the NVD entry and the linked patch and release notes to confirm which document sources trigger the prompt injection and what actions the agent can take.","how_to_check":"- Check the installed version of DeepSeek-Reasonix; versions up to `2.52.0` are affected.\n- Look at whether the tool reads project documents through `internal/memory/doc.go` (`discoverDocs`) in your build.","plain_action":"","needs_analyst":true,"not_affected_if":["version-2.53.0-or-later"],"who_is_affected":"Teams running DeepSeek-Reasonix up to version 2.52.0 where the tool discovers and feeds documents into the model prompt. The sources do not say whether a default configuration is affected or what the document sources are, so whether a given setup is exposed is unknown.","user_interaction":"unknown","needs_special_setup":"unknown","severity_if_unknown":"high"},"cost_usd":0.000738,"input_tokens":4356,"output_tokens":604,"severity_before":"high"},"analyst_status":null,"analyst_note":null,"search":"'-123653':16A,19A '-2026':15A,18A '2.52.0':88C '2.53.0':129C '4ca33493a673c2788aa8598fbca2b2f47a69e1b8':142C 'abl':131C 'address':133C 'affect':89C,149C 'ai':21B,25B,37B 'ai-coding-tool':20B 'ai-tool':24B 'attack':10A,62B,111C 'code':22B,38B 'compon':150C 'content':50B 'craft':49B 'cve':14A,17A 'deepseek':2A,28B,34B,84C 'deepseek-reasonix':1A,27B,33B,83C 'detect':80C 'disclos':74B,120C 'discoverdoc':46B,93C 'discoveri':8A 'document':7A 'esengin':82C 'exploit':71B,117C 'fail':40B 'file':96C 'function':47B,92C 'identifi':137C 'improp':102C 'initi':114C 'inject':5A,52B 'input':43B,105C 'instruct':53B 'internal/memory/doc.go':97C 'issu':135C 'lead':100C 'let':9A 'llm':31B,56B,108C 'llm-prompt':30B 'manipul':99C 'may':123C 'model':13A 'neutral':42B,103C 'patch':140C 'prompt':4A,32B,57B,109C 'public':70B,121C 'reasonix':3A,29B,35B,85C 'remot':66B,115C 'say':60B 'secur':76C 'sourc':59B 'start':65B 'steer':11A 'suggest':145C 'tool':23B,26B,39B 'upgrad':126C,147C 'use':106C,125C 'version':128C 'vulner':77C","nicknames":[],"mitigations":[],"who_affected":"Teams running DeepSeek-Reasonix up to version 2.52.0 where the tool discovers and feeds documents into the model prompt. The sources do not say whether a default configuration is affected or what the document sources are, so whether a given setup is exposed is unknown.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"available","audience":"developers","plain_action":null,"pkeys":[],"ptokens":[],"labels":["aicodingtools","aitooling","deepseekreasonix","high","llmprompting"],"campaign_type":null,"search_a":"'-123653':16A,19A '-2026':15A,18A 'attack':10A 'cve':14A,17A 'deepseek':2A 'deepseek-reasonix':1A 'discoveri':8A 'document':7A 'inject':5A 'let':9A 'model':13A 'prompt':4A 'reasonix':3A 'steer':11A","search_b":"'ai':21B,25B,37B 'ai-coding-tool':20B 'ai-tool':24B 'attack':62B 'code':22B,38B 'content':50B 'craft':49B 'deepseek':28B,34B 'deepseek-reasonix':27B,33B 'disclos':74B 'discoverdoc':46B 'exploit':71B 'fail':40B 'function':47B 'inject':52B 'input':43B 'instruct':53B 'llm':31B,56B 'llm-prompt':30B 'neutral':42B 'prompt':32B,57B 'public':70B 'reasonix':29B,35B 'remot':66B 'say':60B 'sourc':59B 'start':65B 'tool':23B,26B,39B","search_c":"'2.52.0':88C '2.53.0':129C '4ca33493a673c2788aa8598fbca2b2f47a69e1b8':142C 'abl':131C 'address':133C 'affect':89C,149C 'attack':111C 'compon':150C 'deepseek':84C 'deepseek-reasonix':83C 'detect':80C 'disclos':120C 'discoverdoc':93C 'esengin':82C 'exploit':117C 'file':96C 'function':92C 'identifi':137C 'improp':102C 'initi':114C 'input':105C 'internal/memory/doc.go':97C 'issu':135C 'lead':100C 'llm':108C 'manipul':99C 'may':123C 'neutral':103C 'patch':140C 'prompt':109C 'public':121C 'reasonix':85C 'remot':115C 'secur':76C 'suggest':145C 'upgrad':126C,147C 'use':106C,125C 'version':128C 'vulner':77C","first_public_at":"2026-10-27T03:24:01+00:00","kev":null,"cvss_vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:L/VI:L/VA:L/SC:N/SI:N/SA:N/E:P/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","cvss_by":"cna@vuldb.com","poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-123653","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-123653","severity":null,"published_at":"2026-11-06T14:17:10+00:00","modified_at":"2026-11-06T14:17:11+00:00","first_seen_at":"2026-11-06T14:42:11+00:00","fetched_at":"2026-11-06T14:42:11+00:00","note":null}],"documents":[{"id":87033,"url":"https://github.com/esengine/DeepSeek-Reasonix/commit/4ca33493a673c2788aa8598fbca2b2f47a69e1b8","title":"Patch: github.com","kind":"other","source":"nvd","publisher":"vendor","text":null,"published_at":null,"relation":"fix","confidence":null,"note":"Patch: github.com"},{"id":87034,"url":"https://github.com/esengine/DeepSeek-Reasonix/releases/tag/v2.53.0","title":"Release notes: github.com","kind":"release","source":"nvd","publisher":"vendor","text":null,"published_at":null,"relation":"fix","confidence":null,"note":"Release notes: github.com"},{"id":87035,"url":"https://github.com/esengine/DeepSeek-Reasonix/security/advisories/GHSA-7v48-x992-r5w7","title":"Vendor advisory: github.com","kind":"vendor","source":"nvd","publisher":"vendor","text":null,"published_at":"2026-10-27T03:24:01+00:00","relation":"advisory","confidence":null,"note":"Vendor advisory: github.com"}],"exploit_sources":[],"notice":"Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai"}