{"id":2817,"date":"2026-11-10T11:22:37","date_gmt":"2026-11-10T11:22:37","guid":{"rendered":"https:\/\/exeport.eu\/sk\/?p=2817"},"modified":"2026-11-10T11:22:38","modified_gmt":"2026-11-10T11:22:38","slug":"je-pri-praci-s-ai-najvacsim-rizikom-hacker","status":"publish","type":"post","link":"https:\/\/exeport.eu\/sk\/blog\/je-pri-praci-s-ai-najvacsim-rizikom-hacker\/","title":{"rendered":"Je pri pr\u00e1ci s AI najv\u00e4\u010d\u0161\u00edm rizikom hacker?"},"content":{"rendered":"\n

Va\u0161i zamestnanci u\u017e AI pou\u017e\u00edvaj\u00fa.
Skuto\u010dnou ot\u00e1zkou je, kam pri tom putuj\u00fa va\u0161e \u00fadaje,
kto k nim nesk\u00f4r z\u00edska pr\u00edstup
a \u010di ich dok\u00e1\u017eete ochr\u00e1ni\u0165.<\/em><\/p>\n\n\n\n

Kolega pod tlakom term\u00ednu<\/strong><\/p>\n\n\n\n

\u010co dnes v\u00e1\u0161 t\u00edm vlo\u017eil do verejn\u00e9ho AI chatbota?<\/p>\n\n\n\n

Pre v\u00e4\u010d\u0161inu firiem nie je najv\u00e4\u010d\u0161\u00edm rizikom AI hacker. Je n\u00edm kolega, ktor\u00fd to mysl\u00ed dobre, no tla\u010d\u00ed ho term\u00edn, a tak vlo\u017e\u00ed zmluvu, tendrov\u00fa dokument\u00e1ciu, zdrojov\u00fd k\u00f3d alebo klientsky s\u00fabor do verejn\u00e9ho chatbota, aby pr\u00e1cu stihol r\u00fdchlej\u0161ie.<\/p>\n\n\n\n

Samsung sa o tom presved\u010dil na vlastnej ko\u017ei. V marci 2023, len nieko\u013eko t\u00fd\u017ed\u0148ov po tom, \u010do jeho div\u00edzia polovodi\u010dov povolila ChatGPT, do\u0148 in\u017einieri vlo\u017eili intern\u00fd zdrojov\u00fd k\u00f3d a obsah por\u00e1d. Firma potom 1. m\u00e1ja 2023 obmedzila pou\u017e\u00edvanie generat\u00edvnej AI na svojich zariadeniach a v sie\u0165ach (Bloomberg<\/a>; \u010dasov\u00e1 os<\/a>).<\/p>\n\n\n\n

\n\n\n\n

\u010co sa stane s \u00fadajmi vlo\u017een\u00fdmi do verejn\u00e9ho chatbota<\/strong><\/p>\n\n\n\n

Text sa v okamihu odoslania ulo\u017e\u00ed na servery prev\u00e1dzkovate\u013ea. T\u00fdm sa e\u0161te automaticky nest\u00e1va verejn\u00fdm. Va\u0161a firma v\u0161ak nad n\u00edm str\u00e1ca kontrolu a u\u017e ho nedok\u00e1\u017ee vzia\u0165 sp\u00e4\u0165.<\/p>\n\n\n\n

\u00dadaje s\u00fa odteraz u tretej strany, podliehaj\u00fa jej podmienkam a jej jurisdikcii. Dosta\u0165 sa k nim m\u00f4\u017eu viacer\u00e9 skupiny a vy sa o tom v\u00f4bec nemus\u00edte dozvedie\u0165:<\/p>\n\n\n\n

\u2013 Zamestnanci a extern\u00ed dod\u00e1vatelia prev\u00e1dzkovate\u013ea.<\/strong> Google uv\u00e1dza, \u017ee \u013eudsk\u00ed hodnotitelia \u010d\u00edtaj\u00fa a anotuj\u00fa konverz\u00e1cie z aplik\u00e1ci\u00ed Gemini, aby zlep\u0161ovali jeho produkty, a \u017eiada pou\u017e\u00edvate\u013eov, aby nezad\u00e1vali d\u00f4vern\u00e9 inform\u00e1cie (Gemini Apps Privacy Hub<\/a>).<\/p>\n\n\n\n

\u2013 S\u00fady.<\/strong> V spore s denn\u00edkom New York Times dostalo OpenAI v roku 2025 pr\u00edkaz uchov\u00e1va\u0165 logy ChatGPT vr\u00e1tane vymazan\u00fdch chatov. Pr\u00edkaz bol nesk\u00f4r zru\u0161en\u00fd, ale u\u017e uchovan\u00e9 logy zost\u00e1vaj\u00fa stran\u00e1m sporu k dispoz\u00edcii (Yahoo News<\/a>).<\/p>\n\n\n\n

\u2013 \u00dato\u010dn\u00edci s ukradnut\u00fdm \u00fa\u010dtom.<\/strong> Spolo\u010dnos\u0165 Group-IB na\u0161la viac ako 100 000 prihlasovac\u00edch \u00fadajov k ChatGPT v logoch malv\u00e9ru na kr\u00e1de\u017e \u00fadajov (infostealer), s ktor\u00fdmi sa obchoduje na darkwebe. Kto sa do \u00fa\u010dtu prihl\u00e1si, vid\u00ed cel\u00fa hist\u00f3riu konverz\u00e1ci\u00ed (TechRadar<\/a>).<\/p>\n\n\n\n

\u2013 \u0160kodliv\u00e9 roz\u0161\u00edrenia prehliada\u010da.<\/strong> V roku 2025 v\u00fdskumn\u00edci zistili, \u017ee Urban VPN a s\u00favisiace roz\u0161\u00edrenia pre Chrome a Edge zbierali kompletn\u00e9 konverz\u00e1cie s AI od viac ako 8 mili\u00f3nov pou\u017e\u00edvate\u013eov a pred\u00e1vali ich d\u00e1tov\u00fdm brokerom (The Register<\/a>).<\/p>\n\n\n\n

\u2013 Ktoko\u013evek s vyh\u013ead\u00e1va\u010dom.<\/strong> V polovici roka 2025 sa zdie\u013ean\u00e9 konverz\u00e1cie z ChatGPT ozna\u010den\u00e9 ako doh\u013eadate\u013en\u00e9 objavovali vo v\u00fdsledkoch vyh\u013ead\u00e1vania Google, k\u00fdm OpenAI t\u00fato funkciu nezru\u0161ilo (Fast Company<\/a>).<\/p>\n\n\n\n

\n\n\n\n

Ako reaguj\u00fa popredn\u00e9 firmy<\/strong><\/p>\n\n\n\n

Ve\u013ek\u00e9 organiz\u00e1cie pre\u0161li tromi f\u00e1zami a mnoh\u00e9 dnes uplat\u0148uj\u00fa v\u0161etky tri naraz.<\/p>\n\n\n\n

1. Plo\u0161n\u00e9 z\u00e1kazy.<\/strong> Vo febru\u00e1ri 2023 JPMorgan Chase, Bank of America, Citigroup, Deutsche Bank, Goldman Sachs a Wells Fargo svojim zamestnancom obmedzili alebo zablokovali ChatGPT (Fortune<\/a>; Bloomberg<\/a>).<\/p>\n\n\n\n

2. Kontrolovan\u00fd pr\u00edstup k schv\u00e1len\u00fdm n\u00e1strojom.<\/strong> Firmy zav\u00e1dzaj\u00fa podnikov\u00e9 verzie AI asistentov so zmluvami, ktor\u00e9 vylu\u010duj\u00fa tr\u00e9novanie na ich \u00fadajoch, a kombinuj\u00fa ich s n\u00e1strojmi na prevenciu straty \u00fadajov (DLP), ktor\u00e9 kontroluj\u00fa, \u010do zamestnanci vkladaj\u00fa.<\/p>\n\n\n\n

3. Vlastn\u00e1 AI pre najcitlivej\u0161ie \u00fadaje.<\/strong> Samsung po \u00faniku za\u010dal vyv\u00edja\u0165 vlastn\u00e9 intern\u00e9 n\u00e1stroje AI (Bloomberg<\/a>). Obrann\u00e9 a spravodajsk\u00e9 organiz\u00e1cie id\u00fa e\u0161te \u010falej a pou\u017e\u00edvaj\u00fa AI, ktor\u00e1 be\u017e\u00ed v \u00faplne izolovanom re\u017eime (air-gapped), bez pripojenia na internet (Nextgov<\/a>).<\/p>\n\n\n\n

\n\n\n\n

Pre\u010do samotn\u00e9 z\u00e1kazy nefunguj\u00fa<\/strong><\/p>\n\n\n\n

Z\u00e1kaz AI v pr\u00e1ci nezabr\u00e1ni \u013eu\u010fom, aby ju pou\u017e\u00edvali. Len v\u00e1m znemo\u017en\u00ed vidie\u0165, \u010do do nej vkladaj\u00fa.<\/p>\n\n\n\n

\u010c\u00edsla to potvrdzuj\u00fa. Pod\u013ea spr\u00e1vy Verizon 2026 Data Breach Investigations Report dnes 45 % zamestnancov pravidelne pou\u017e\u00edva AI na firemn\u00fdch zariadeniach, k\u00fdm pred rokom to bolo 15 %, a 67 % pou\u017e\u00edvate\u013eov sa na nich k slu\u017eb\u00e1m AI prip\u00e1ja cez nefiremn\u00e9 \u00fa\u010dty. Naj\u010dastej\u0161\u00edm typom \u00fadajov nahr\u00e1van\u00fdch do neschv\u00e1len\u00fdch n\u00e1strojov AI bol zdrojov\u00fd k\u00f3d (Help Net Security<\/a>; Mimecast<\/a>).<\/p>\n\n\n\n

D\u00f4sledky sa prejavuj\u00fa aj v \u0161tatistik\u00e1ch \u00fanikov. Pod\u013ea spr\u00e1vy IBM Cost of a Data Breach Report 2026 nahl\u00e1silo 43 % sk\u00faman\u00fdch organiz\u00e1ci\u00ed, ktor\u00e9 za\u017eili \u00fanik \u00fadajov, bezpe\u010dnostn\u00fd incident spojen\u00fd s tie\u0148ovou AI (shadow AI), k\u00fdm pred rokom to bolo 20 %. Tak\u00e9to \u00faniky st\u00e1li v priemere 5,39 mili\u00f3na USD (IBM Newsroom<\/a>; Kiteworks<\/a>).<\/p>\n\n\n\n

\n\n\n\n

Ani platformy AI nie s\u00fa nedotknute\u013en\u00e9<\/strong><\/p>\n\n\n\n

\u00dadaje odoslan\u00e9 mimo va\u0161ej siete kon\u010dia na infra\u0161trukt\u00fare, ktor\u00fa nevlastn\u00edte ani nekontrolujete. A prelomi\u0165 sa daj\u00fa aj platformy v samom srdci odvetvia AI.<\/p>\n\n\n\n

V j\u00fali 2026 agenti AI, ktor\u00fdch OpenAI testovalo, unikli z testovacieho prostredia a prenikli do produk\u010dnej infra\u0161trukt\u00fary Hugging Face. Nikto im to neprik\u00e1zal. Sna\u017eili sa podv\u00e1dza\u0165 pri teste t\u00fdm, \u017ee h\u013eadali odpovede na internete. OpenAI potvrdilo, \u017ee bezpe\u010dnostn\u00e9 poistky modelov boli na \u00fa\u010dely testu z\u00e1merne zn\u00ed\u017een\u00e9 (CNBC<\/a>).<\/p>\n\n\n\n

\n\n\n\n

AI za m\u00farmi va\u0161ej firmy<\/strong><\/p>\n\n\n\n

Ot\u00e1zkou u\u017e nie je, \u010di v\u00e1\u0161 t\u00edm AI pou\u017e\u00edva, ale kam pri tom putuj\u00fa va\u0161e \u00fadaje. Sentinea ich ponech\u00e1 vo vlastnej infra\u0161trukt\u00fare.<\/p>\n\n\n\n

Sentinea je platforma AI s d\u00f4razom na bezpe\u010dnos\u0165, ktor\u00e1 be\u017e\u00ed priamo u v\u00e1s:<\/p>\n\n\n\n

\u2013 Prekladajte d\u00f4vern\u00e9 dokumenty<\/strong> bez toho, aby ste ich posielali externej slu\u017ebe.<\/p>\n\n\n\n

\u2013 Z\u00edskajte odpovede z intern\u00fdch znalost\u00ed<\/strong> na z\u00e1klade vlastn\u00fdch dokumentov.<\/p>\n\n\n\n

\u2013 Prev\u00e1dzkujte ju on-premises alebo v \u00faplne izolovanom re\u017eime (air-gapped)<\/strong>, v najcitlivej\u0161\u00edch prostrediach \u00faplne bez pripojenia na internet.<\/p>\n\n\n\n

\u2013 Ria\u010fte pr\u00edstup<\/strong> pomocou opr\u00e1vnen\u00ed pod\u013ea rol\u00ed, jednotn\u00e9ho prihl\u00e1senia (SSO) a auditn\u00fdch z\u00e1znamov.<\/p>\n\n\n\n

Sentinea nenahr\u00e1dza va\u0161e intern\u00e9 pravidl\u00e1 ani DLP. Odstra\u0148uje v\u0161ak d\u00f4vod ich poru\u0161ova\u0165: zamestnanci dostan\u00fa bezpe\u010dn\u00fa alternat\u00edvu, ktor\u00e1 je rovnako pohodln\u00e1 ako verejn\u00e1 AI.<\/p>\n\n\n\n

Your On-Premises AI. Zero Off-Premises Exposure.<\/strong><\/p>\n\n\n\n

<\/div>\n\n\n\n

Viac inform\u00e1ci\u00ed n\u00e1jdete na <\/strong>sentinea.com<\/span><\/strong><\/a>.<\/strong><\/p>\n\n\n\n

<\/div>\n\n\n\n

Zdroje<\/strong><\/p>\n\n\n\n