{"id":147923,"date":"2025-11-10T23:11:54","date_gmt":"2025-11-10T21:11:56","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/ataka-whisper-leak-po-opredeleniyu-tematiki-obshheniya-s-ai-chatbotami-cherez-analiz-tls-trafika"},"modified":"2025-11-10T23:11:54","modified_gmt":"2025-11-10T21:11:56","slug":"ataka-whisper-leak-po-opredeleniyu-tematiki-obshheniya-s-ai-chatbotami-cherez-analiz-tls-trafika","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/novosti-interneta\/ataka-whisper-leak-po-opredeleniyu-tematiki-obshheniya-s-ai-chatbotami-cherez-analiz-tls-trafika","title":{"rendered":"Sulmi Whisper Leak p\u00ebr p\u00ebrcaktimin e tematik\u00ebs s\u00eb komunikimit me AI-chatbot\u00ebt p\u00ebrmes analiz\u00ebs s\u00eb trafikut TLS","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Hulumtuesit e siguris\u00eb nga kompania Microsoft zhvilluan nj\u00eb teknik\u00eb sulmi p\u00ebrmes kanaleve jonormale t\u00eb quajtur Whisper Leak, e cila lejon klasifikimin e temave t\u00eb k\u00ebrkesave p\u00ebr sh\u00ebrbimet e bazuara n\u00eb modelet e m\u00ebdha gjuh\u00ebsore p\u00ebrmes nj\u00eb analize passive t\u00eb trafikut t\u00eb enkriptuar t\u00eb transmetuar p\u00ebrmes nj\u00eb lidhjeje TLS. Informacioni mbi madh\u00ebsin\u00eb e paketave rrjet dhe vonesat midis transmetimeve u tregua mjaftuesh\u00ebm p\u00ebr t\u00eb p\u00ebrcaktuar temat e k\u00ebrkesave p\u00ebr chatbot\u00ebt AI me sakt\u00ebsi mbi 98%. Praktikisht, metoda e propozuar mund t\u00eb p\u00ebrdoret p\u00ebr t\u00eb identifikuar n\u00eb trafikun transit t\u00eb p\u00ebrdoruesve tema t\u00eb caktuara k\u00ebrkesash, p\u00ebr shembull, p\u00ebrpjekje p\u00ebr t\u00eb marr\u00eb informacion mbi veprimtari t\u00eb paligjshme, pa dekoduar p\u00ebrmbajtjen. <\/p>\n
Vegla p\u00ebr nxjerrjen e t\u00eb dh\u00ebnave nga grumbujt e trafikut, trajnim i modelit dhe testimi i funksionimit t\u00eb metod\u00ebs \u00ebsht\u00eb publikuar n\u00eb GitHub. Mund\u00ebsia e realizimit t\u00eb nj\u00eb sulmi \u00ebsht\u00eb demonstruar p\u00ebr 28 modele t\u00eb njohura t\u00eb gjuh\u00ebs s\u00eb madhe nga prodhuesit m\u00eb t\u00eb m\u00ebdhenj. P\u00ebr shembull, sakt\u00ebsia e identifikimit t\u00eb k\u00ebrkesave mbi tem\u00ebn \"pastrimi i parave\" p\u00ebr shum\u00eb sh\u00ebrbime AI ishte 100%, n\u00eb prani t\u00eb nj\u00eb k\u00ebrkese t\u00eb d\u00ebshiruar n\u00eb trafikun e analizuar dhe 10,000 k\u00ebrkesa q\u00eb nuk haseshin me tem\u00ebn e nevojshme.
<\/center> <\/p>\n
Arsyeja e shkarkimit t\u00eb informacionit \u00ebsht\u00eb se modelet gjenerojn\u00eb p\u00ebrgjigje p\u00ebr k\u00ebrkesat hap pas hapi, nj\u00eb token n\u00eb \u00e7do her\u00eb, duke p\u00ebrdorur tokenin e m\u00ebparsh\u00ebm si kontekst p\u00ebr t\u00eb p\u00ebrcaktuar fjal\u00ebn ose fraz\u00ebn m\u00eb t\u00eb mundshme t\u00eb ardhshme. Sipas k\u00ebsaj, p\u00ebr \u00e7do token d\u00ebrgohet nj\u00eb paket\u00eb rrjeti e ve\u00e7ant\u00eb dhe vonesa nd\u00ebrmjet paketave korrespondon me vones\u00ebn nd\u00ebrmjet p\u00ebrcaktimit t\u00eb tokenit t\u00eb ardhsh\u00ebm nga modeli. <\/p>\n
N\u00eb TLS, n\u00ebse s'ka kompresim t\u00eb dh\u00ebnash, shk\u00ebmbimi i tekstit t\u00eb enkriptuar barazon me madh\u00ebsin\u00eb e tekstit t\u00eb pa enkriptuar plus nj\u00eb konstant\u00eb. Kur krijohet nj\u00eb model q\u00eb mapon grupe t\u00eb k\u00ebrkuara tokenesh me madh\u00ebsit\u00eb e paketave dhe vonesat nd\u00ebrmjet d\u00ebrgimit t\u00eb tyre, \u00ebsht\u00eb e mundur t\u00eb p\u00ebrcaktohet mjaft sakt\u00eb prania e temave t\u00eb k\u00ebrkuara n\u00eb trafik. N\u00eb hulumtim jan\u00eb p\u00ebrgatitur tri variante t\u00eb modeleve t\u00eb tilla t\u00eb m\u00ebsimit t\u00eb makineris\u00eb, t\u00eb bazuara n\u00eb arkitekturat e rrjeteve nervore LightGBM, Bi-LSTM dhe BERT. P\u00ebr \u00e7do model jan\u00eb kryer eksperimentet p\u00ebr t\u00eb p\u00ebrcaktuar tem\u00ebn e k\u00ebrkuar gjat\u00eb analiz\u00ebs duke shqyrtuar vet\u00ebm madh\u00ebsin\u00eb e pakove, vet\u00ebm vonesat midis pakove dhe t\u00eb dy kriteret.
![\"Sulmi](\"\/wp-content\/uploads\/2025\/11\/9e2db8a564ae07b2ae7ced2d2f744e46.png\")
<\/center> <\/p>\n![\"Sulmi](\"\/wp-content\/uploads\/2025\/11\/d6be536d5b8814af45a42317c9454a44.png\")
<\/center> <\/p>\n