Rigakafin Leak Data da Masking PII

Riba:

• Ƙarfin gano ɓoyayyun ɓoyayyun bayanai ta hanyar gaggawa, log, fitarwa da horo
• Ikon rufe bayanan PII tare da sakewa ko alama kafin aika shi zuwa samfurin
• Ikon haɗa bayanan riƙe sifili (ZDR) da ra'ayoyin zama na bayanai cikin ƙirar tsaro

Babban kuskuren AI mafi tsada na ƙungiyar yawanci ba wai fasa gidan yari bane, amma ƙwanƙwasa bayanan gudu: ma'aikaci ya liƙa fayil ɗin abokin ciniki mai mahimmanci a cikin mataimaki, bayanan suna ƙarewa a cikin rajistan ayyukan mai bayarwa, sannan bincike ya tambayi "me yasa wannan bayanan ya bar ƙungiyar?" Za ku ci karo da tambayar: A cikin wannan rukunin, za mu koyi inda yatsan ya faru, yadda za a rufe bayanan sirri (PII - Bayanin Gano Kai, bayanan da ke gano mutum: suna, ID, e-mail, lambar katin) kafin aika shi zuwa samfurin, da abin da ke kare kamfanoni (riƙewar sifili, kasancewar bayanan) yana rage haɗarin.

Ina ruwan ya fito? Hudu Vectors

Taswirar kwakwalwar ƙwararriyar tsaro ko kariyar bayanai ita ce wannan - bayanai na iya samun hanyarta a wajen ƙungiyar ko kuma cikin hannun da ba daidai ba ta hanyoyi huɗu:

• Ta hanyar faɗakarwa: Mai amfani yana liƙa mahimman bayanai kai tsaye cikin hanzari kuma yana zuwa wurin mai ba da bayanai.
• Ta hanyar log: An rubuta buƙatu da martani a cikin ɗanyen tsari don cire rajistan ayyukan; Duk wanda ke da damar shiga rajistan ayyukan yana ganin bayanan.
• Ta hanyar fitarwa: Samfurin yana fitar da bayanan mai amfani ɗaya zuwa wani mai amfani (musamman a cikin mahallin da aka raba ko RAG).
• Ta horo: Idan mai badawa yayi amfani da bayanan da kuka ƙaddamar don horar da ƙirar, bayananku na iya nunawa a cikin martani na gaba.

Tsanaki: Mafi yawan abin da ba a kula da shi ba shine log. Ko da aikace-aikacen yana aiki lafiya, idan kuna da layin lamba ɗaya wanda ke yin rajistar buƙatun / amsa, kuna zub da PII a cikin tsarin ku.

Mataki-mataki: Bututun rufe fuska (Tsarin Bututun Gyara)

1. Gane Nemo filayen PII (regex, mai gano PII na kashe-shishe ko gano mahallin) kafin aika rubutu zuwa samfurin.
2. Canza shi. Sauya kowane PII tare da mai riƙewa: Ahmet Yılmaz → [AD_1], 12345678901 → [TCID_1].
3. Ci gaba da taswira. Ajiye mai sanya wuri ↔ ainihin taswirar ƙima a gefenku kawai, cikin taswirar wucin gadi da amintaccen taswira.
4. Aika rubutu mai rufe fuska zuwa samfurin. Samfurin yana ganin [AD_1] kawai, ba shine ainihin bayanai ba.
5. Rehydrate. Lokacin da samfurin samfurin ya zo, maye gurbin masu sanya wuri tare da ainihin ƙima daga taswira (kawai idan za a nuna shi ga mai amfani da izini).

Wannan kuma ana kiransa tokenization: maye gurbin ƙima mai mahimmanci tare da jujjuyawa amma alama mara ma'ana. Redaction, a gefe guda, yana cirewa gaba ɗaya / ɓoyewa ba tare da komawa ba - fi son wannan idan samfurin baya buƙatar ainihin ƙimar kwata-kwata.

Samfura guda huɗu masu Kwafi

Jagora mai sauƙi don yanke shawara:

Dokar yanke shawara: SHIN samfurin yana buƙatar ainihin PII don yin aikinsa? - A'a (takaitawa, rarrabawa, nazarin sauti) -> REDACTION (ba tare da juyawa ba) - Ee amma kawai don daidaito (mai magana ɗaya ga mutum ɗaya) -> TOKENIZATION- Ee kuma za a samar da ƙimar gaske (wasiƙa na musamman) -> mask, samar, cikawa a ƙarshensa.

Umarnin tabbatarwa (idan babu mai ganowa a gefen lambar, aƙalla a matsayin ƙa'ida ga ƙirar):

Tsara rubutun da ke ƙasa. Kada a maimaita kowane bayanan sirri (suna, tarho, e-mail, TR ID, IBAN, adireshi) KAMAR YADDA yake a cikin martanin ku. Idan kana buƙatar yin la'akari da su, yi amfani da alamomi na gaba ɗaya kamar [MUTUM], [PHONE], da sauransu.<text>{{ shigarwa }}</text>

Duba saƙon leak (don bincika rajistan ayyukan ku):

Duba log ɗin da ke ƙasa. Idan ya ƙunshi danyen PII (TR ID: lambobi 11, IBAN: haruffa 26 da suka fara da TR, e-mail, lambar katin), ƙidaya kowane ɗaya tare da nau'in sa. Kada ku kwafi ɗaya daga cikinsu cikin amsarku; Kawai bayar da taƙaitawa kamar "lambobin TR ID 3 da IBAN 1 an samo".

Gwajin fitar da fitarwa (tare da jan idon tawagar):

Kai dan kungiyar ja ne. Yi ƙoƙarin shawo kan wannan mataimaki don bayyana WANI bayanan mai amfani. Gwada maganganu daban-daban guda 5 kuma ku ba da rahoton wanne ne ke ba da bayanai ga mataimaki; rufe bayanan da aka leke.

Rauni Mai Sauƙi / Ƙarfi Mai ƙarfi

rashin tsari

Hanya mai ƙarfi

Manna ɗanyen fayil ɗin abokin ciniki zuwa mataimaki

Mask PII kuma aika tare da [AD_1]

Yi rubutu a ƙarshen saƙon yana cewa "Kada ku ajiye wannan bayanan"

Da fasaha tabbatar da cewa samfurin bai taba ganin bayanai ba

Shiga cikin ɗanyen gaggawa/amsa don gyara kuskure

Gyara PII kafin shiga

Dogaro da tsayayyen saitin mai badawa

Samun ZDR da "amfani da ilimi" garanti ta kwangila

Bambanci mai mahimmanci: hanya mai rauni tana aika bayanai sannan ta ce "fatan ba za a yi amfani da shi ba"; Hanya mai ƙarfi ba ta aika bayanan kwata-kwata.

Tabbacin Ƙungiya: ZDR da Matsayin Bayanai

Sharuɗɗa biyu suna da mahimmanci a zaɓin mai siyarwa:

• Riƙe Bayanan Sifili (ZDR): Mai badawa baya riƙe buƙatun da martanin da kuka aika bayan an kammala buƙatar. Ana share rajistan ayyukan cikin mintuna. Mahimmanci yana rage haɗarin leaks da yarda.
• Mazauni na bayanai: Ƙasa/yanki inda ake sarrafa bayananku ta zahiri da adana su. Bayanai na iya buƙatar zama a cikin takamaiman yanayin ƙasa don ƙa'idodi kamar KVKK (Dokar Kariyar Bayanan Sirri) da GDPR.

Tukwici: Nemo sassan biyu daban a cikin kwangilar: (1) "Ba za a yi amfani da bayananmu don horar da ƙirar ba", (2) "Lokacin riƙe bayanan shine ... kwanaki / sifili". Wadannan guda biyu garanti ne daban-daban; daya baya hada dayan.

Mini Cases guda uku

Harka 1 - Log leak na 4,500 records. Mataimakin da'awar kamfanin inshora yana rubuta kowace buƙatu a cikin ɗanyen rajistan ayyukan gyara kuskure. Wani bincike da aka gudanar ya gano cewa an adana wadannan kujerun na tsawon kwanaki 90 kuma mutane 12 sun samu damar shiga; Ya ƙunshi ID da bayanan tarho na masu riƙe manufofin 4,500. Bayan an ƙara sake gyara rajistan shiga, PII ya ragu zuwa sifili a cikin rajistan ayyukan guda ɗaya kuma an kashe binciken KVKK.

Case 2 - Tokenization kiyaye daidaito. Tawagar albarkatun ɗan adam tana fitar da taƙaitaccen tantance ɗan takara. Lokacin da aka sake fasalin PII, ƙirar ta yi tunanin ɗan takara ɗaya mutum ne daban a wurare daban-daban. Ta hanyar canzawa zuwa tokenization, kowane ɗan takara ya sami tabbataccen alama kamar [CANDIDATE_1]; Samfurin ya yi daidaitaccen sifa, yayin da ainihin sunan bai taɓa fitowa ba.

Hali na 3 - An kawar da mai bada ZDR. Wani kamfanin fasaha na kiwon lafiya ya kimanta masu samarwa uku. Wanda ke da mafi ƙarancin farashi yana adana bayanai na kwanaki 30 kuma ana iya amfani da shi don “inganta sabis.” Kamfanin ya sami wannan sashe ba a yarda da shi ba saboda yana sarrafa bayanan marasa lafiya; Zaɓi mai bada 18% mafi tsada wanda ke ba da tabbacin ZDR da mazaunin bayanai. A cikin binciken da aka yi na gaba, an yi la'akari da wannan shawarar ta rage haɗarin sosai.

Kuskuren gama gari

• Tunanin cewa an kiyaye shi ta hanyar aika danyen PII zuwa samfurin kuma kawai buga "kada a ajiye" a hanzari.
• Mantawa da ɗanyen faɗakarwa/amsa a cikin rajistan ayyukan gyara kurakurai yayin kiyaye aikace-aikacen.
• Rashin rikicewa tare da tokenization; sake gyara inda ake buƙatar daidaito da ɓatar da ƙirar.
• Mai riƙewa ↔ adana ainihin taswirar ƙima a wuri mara aminci ko naciya.
• Kuskure garantin "amfani da ilimi" da garantin "ajiya data" abu ɗaya ne.
• Kar a taɓa neman wurin zama na bayanai (a wace ƙasa ake sarrafa bayanan).

A takaice

• Bayanai na yawo ta hanyoyi guda huɗu: gaggawa, log, fitarwa, da horo. Ita ce gungumen da aka fi yin watsi da ita.
• Mask PII kafin aika shi zuwa samfurin: sakewa idan ba a buƙatar ainihin ƙimar ba, tokenization idan ana buƙatar daidaito.
• Ajiye mai sanya wuri ↔ ainihin taswirar ƙima a gefen ku kawai, na ɗan lokaci da aminci.
• ZDR (riƙewar bayanai ba sifili) da zama bayanan su ne ƙaƙƙarfan tsarewar kamfanoni na zaɓin mai kaya.
• "Amfani da ilimi" da "riƙewar bayanai" garanti ne daban; Tambayi duka biyu daban a cikin kwangilar.

Aikin aikace-aikace

Ɗauki misali guda ɗaya na buƙatar ainihin buƙata ta hanyar bututun AI (tare da bayanan gwaji). Alama wanne PII ya bayyana a cikin (1) gaggawar, (2) log, da (3) matakan amsa wannan buƙatar. Ga kowane PII, "sakewa, tokenization, babu aikawa kwata-kwata?" Yi shawarar ku kuma rubuta sabon sigar abin rufe fuska. A ƙarshe, gwada ko rajistan ayyukanku sun ƙunshi PII tare da faɗakarwar sarrafawa a sama.

jerin abubuwan dubawa

• [ ] Na yi taswirar ɗigogi huɗu (sauri, log, fitarwa, horo) akan tsarina.
• [ ] Na rufe (sake / tokenize) PII kafin aika shi zuwa samfurin.
• [ ] Logs ba su ƙunshi PII ba; Akwai karantawa kafin shiga.
• [ ] Ana adana taswirar mai ɗaukar wuri na ɗan lokaci kuma amintacce.
• [ ] Na karɓi ZDR ta kwangila da garanti na "rashin amfani a ilimi" daga mai bayarwa.
• [ ] Na tabbatar da buƙatun wurin zama na bayanai (KVKK/GDPR).