LangChain CVE-2025-68664. Ինչpés PII-ě Artahosoum É Yor RAG Pipeline-its

LangChain CVE-2025-68664. Ինchpés PII-ě Artahosoum É Yor RAG Pipeline-its

Tharmaratsvats 2026 th.-i hamar:

2025 th. verjoum LangChain-oum karevоr thoutoutyoun é haytnabervel: CVE-n é CVE-2025-68664: CVSS-i arkhě é 9.3 (Karevоr):

Ayn ouseytsvats é LangChain-i serializatsiayi kodi:

Inch É Anoum CVE-2025-68664-ě

LangChain-ě ouni yerkou serializatsiayi funktsia: dumps() ev dumpd(): Nrank Python-i objektnerě teksti en verchakoum:

Thoutoutyouně closure-i karavaman mej é:

Yerb LangChain-ě serializatsnoum é callable-ě, ayn gravоum é closure-i kontekstě:

Hartakoghy, vory karavоum é LLM-i pataskhannerě, karogh é gontskorel dumps()-ě: Funksian aynouhétev kardum é Python-i protsesí sharchakani pokhokhakannery:

Ardzouchě tvylalneri ennarkoumn é: API banalinnerě, tvylalneri bazayi tоgherě, JWT gaghtniqnerě ev AWS havatoumnerě karogh en haytnovel modeli ardzouchоum:

Hartakoghy, vory tekst é ennarkoum RAG aghbyourayin vastateghti mej, karogh é kardalal yor artadrakan gaghtniqnerě:

Azttvats tarberkannery: LangChain 0.3.22-its pats (Python): 0.3.22 tarberaky ounі ughtоumě:

PyPI-i tvylalnery tsoyts en taroum hin tarberkanerim la tsak ogtagordzoum minchev 2026 th. mart:

Inchpés PII-ě Artahosoum É RAG Pipeline-neroum

CVE-2025-68664-ě dramatik é: Bayts da miayn mek depk é avier layin khndrí:

Tvylalnery RAG pipeline-neri midjov artahosoum en sovorobar: Hartakoghi karik chka:

Aha standart korooratyiv RAG kargavouroum:

Nakhqan, ennerkoum: Kentronakan vastatéghtery indeksavoroum es vector store-i mej: Mtatse support tickets-ner, haghordakhí el. nambaknery, paymanagirery ev HR grakhoumnery:

Taradzqvats vector store-nerě Pinecone-ě, Weaviate-ě ev pgvector-ě en:

Aynouhétev, stanoum: Ogtateré harcоum é: Hamakargě banchagochits hing amenakaptvats khanky stanoum é:

Aveli, steghtoum: Ayd khankery LLM-in — GPT-4o-in, Claude-in kam Gemini-in — en ougharkoum orpés kontekst:

Yerkourrord qayle khndiré é: Staccvats khankery pahоum en ain inchy aghbyourayin vastatéghterě pahоum ein: Da nerenkyelyal:

• Haghordakhí anounner, el. hascery ev herrakhosamamerě
• Paymanagri arzheknerě, hashvapahestí hamarnery ev harkayin nuynakanichnerě
• Ashkhatoghi ashkhatovardzyí tvylalnery ev kataroghoutyan stougebani grakhoumnery
• Hivand anounner klinikakan grakhoumneroum
• Azgayin ID-i hamarnery artagakht fayleroum

Ayd tvylalnery LLM-in en ougharkoum anarjakel: Nrank karogh en haytnovel modeli ardzouchоum:

Nrank LLM matakarari koghmic amragrvоum en: Nstоum en yor zrouytsi patmoushtyyan mej: Hosoum en yor observability stack-i mej:

Hartakoghi karik chka: Sa RAG-i nakhagetsov ashkhatanki dzevn é: Nakhagetsě irakhan gaghtniouyth riski é steghtsоum:

68 Gaghtnyal Dzevacherper Korooratyiv Vastatéghtí Bandzoumneroum

Anvandutyyan gortsiknerě hetevm en 68 haytni gaghtnyal dzevacherpi: Nrank hachakh en haytnoum, qan thimerě spasoum en:

Aha amenatosirachnery:

• AWS Access Key ID-ner (AKIA...)
• OpenAI API banalinnerě (sk-...)
• Anthropic API banalinnerě (sk-ant-...)
• Tvylalneri bazayi URI-ner (postgresql://user:password@host/db)
• JWT token-ner (base64-kodavorvats headers)
• GitHub Personal Access Token-ner
• Stripe-i gaghtnyal banalinnerě (sk_live_...)
• SendGrid API banalinnerě
• Twilio-i hashví SID-nerě ev auth token-nerě
• Masnaworí banalyí PEM bloknery

Support ticket-ě karogh é pahel haghordakhí API banali debug session-its:

Paymanagirě karogh é nerenkyelyal tvylalneri bazayi havatoumner tekhnikakan handoff-its:

Skhalmik indeksavorvats config faylě karogh é batsahatnel amsboghj secrets store-ě:

Yerb ayd faylery vector store-i mej en mтноum aranc sanitization-i, yuraqanchyur query karogh é gaghtniqnerě LLM-in ougharkel:

Nrank karogh en hasnel naev verjin ogtaterum:

Oughtel: Ananounatsel Embedding-its Aradjel

Chistí mote ananounatsnoum é vastatéghtnerě chunking-its ev embedding-its aradjel:

Ayd qaylě parvadír é tsakhayoutyan tvylalner karavaragogh yekhanats hamakargí hamar:

Aha Python-i orinak, ogtagerdzеlov anonym.legal API-ě:

import requests
import os

ANONYM_API_KEY = os.environ["ANONYM_API_KEY"]
ANONYM_BASE_URL = "https://anonym.legal/api"

def anonymize_before_embedding(text: str) -> tuple[str, dict]:
    """Anonymize PII before embedding."""
    response = requests.post(
        f"{ANONYM_BASE_URL}/presidio/anonymize",
        json={
            "text": text,
            "language": "en",
            "anonymizers": {
                "DEFAULT": {"type": "replace", "new_value": "[REDACTED]"},
                "PERSON": {"type": "mask", "masking_char": "*", "chars_to_mask": 4, "from_end": False},
                "EMAIL_ADDRESS": {"type": "replace", "new_value": "[EMAIL]"},
                "PHONE_NUMBER": {"type": "replace", "new_value": "[PHONE]"},
                "CRYPTO": {"type": "replace", "new_value": "[SECRET]"},
                "URL": {"type": "keep"},
            }
        },
        headers={"Authorization": f"Bearer {ANONYM_API_KEY}"}
    )
    result = response.json()
    return result["text"], result.get("items", [])


def build_rag_index(documents: list[str], vectorstore):
    """Build a RAG index with clean documents only."""
    anonymized_docs = []
    for doc in documents:
        clean_text, entities = anonymize_before_embedding(doc)
        anonymized_docs.append(clean_text)
        print(f"Removed {len(entities)} PII entities from document")
    vectorstore.add_texts(anonymized_docs)

anonym.legal API-ě tsadkoum é 285+ kazmakerputyan tesak: Anounner, el. hascery, herrakhosamamerě, azgayin ID-nerě, API banalinnerě ev tvylalneri bazayi URI-nerě borer en brnvоum:

Voch mek zqayoun ban vector store-i chi hasnoum: Ayspes voch mek zqayoun ban ogtaterum artahosvel chi karogh:

Tes tsragravoghi ourkhetsouytse LangChain-i ev LlamaIndex-i kargavouvankyi dzevacherperum:

Oughtel CVE-2025-68664-ě Hima

Ete LangChain-ě 0.3.22-its pats gorkatsnoum es, hima tharmaratsets:

pip install "langchain>=0.3.22" "langchain-core>=0.3.22"

Patching-its heto stougeabanek yor chain configs-ě injection riski hamar: Aha yereq qayl:

Nakhqan, vaveratsrek staccvats chunk-nerě: Arek da nakhqan LLM-in hasnelou:

Hertsel bovanddakoutyouně, vory hamapataskhanoum é injection dzevacherperim, ayspisin orpés ignore previous instructions, system: kam <INST>:

Yerkourrord, ananounatsrek embedding-its aradjel: Da nvazetstnoum é hartaki makersetě:

Ete injection-ě teghi ouni, zqayoun tvylalnery chen lini artahanel:

Yerrerrord, sahmanafetek chain-i tuylatvouyhtounnerě: LangChain chain-nerě petk chi kardalal sharchakani pokhokhakannery avier, qan inchy nrank karik ounén:

Ogtagerdzets minimalakan scope-ov service account:

Matematikan Parzel É

CVSS-i arkhě 9.3 é: Ughtоumě mek API call é mek vastatéghtí hamar:

CVE-2025-68664-i ev hamaynakayin RAG tvylalneri riski kombinatsian irakhan pataskhanattvouyth é:

Loutsouměayts é: Ananounatsir ennerqman zamanin, voch harcasksman:

Stougeabanek anvandutyyan ev hamapataskhanutyyan ambaghchakayin aknarkě korooratyiv RAG patandjneri hamar:

Aghbyurner

• NVD CVE-2025-68664, CVSS 9.3, LangChain serializatsiayi thoutoutyoun
• LangChain anvandutyyan tsanoutsоum, langchain-ai/langchain GitHub, 2025
• OWASP LLM Top 10: LLM01 Prompt Injection, LLM06 Zqayoun Tevekoutyyan Batsaytoum
• anonym.legal kazmakerputyan tesaki vastatéghtеragir — 285+ astitsavats kazmakerputyan tesak