AI bilan bog'liq kiberhujumlar ortib bormoqda, xakerlar AI kirish kalitlarini o'g'irlab, ularni monetizatsiya qilish va hujumlarini tezlashtirish uchun AI vositalaridan foydalanmoqda.
O'tgan oy davomida kamida o'nlab AI ishlab chiquvchi platformalar hujumga uchradi, eng so'nggi hodisalar CommandCode, Northflank, Trigger.dev, Chainstack va WorkOS bilan bog'liq.
Barcha hodisalarda xakerlar monetizatsiya qilish mumkin bo'lgan AI kirish tokenlarini qidirgan, hujumlarni tezlashtirish va soddalashtirish uchun AI agentlaridan foydalangan.
Sanoat miqyosidagi umumiy tendentsiyaning bir qismi bo'lsa-da, hujumlar bir-biri bilan bog'liq emas va dastlabki kirish buzilgan hisob ma'lumotlaridan tortib uchinchi tomon bog'liqliklari va zaifliklardan foydalanishgacha turlicha.
Hujumchilar Andromeda, Brev (NVIDIA), CommandCode, Fireworks AI, Lightning AI, Northflank, OpenCode, Railway, Trigger.dev, Warp va Zeaburga hujum qildi, asosan saqlangan mijoz sirlari qidirildi.
"Buzg'unchilar AI provayderi hisob ma'lumotlariga xosting platformalari, baholash xizmatlari va ishlab chiquvchi konsollari orqali oraliq qatlam orqali yetib bormoqda, va so'nggi shlyuz buzilishlari ham ushbu sirlarga qiziqishni ko'rsatmoqda. Bu qatlam ko'plab pastki mijozlar nomidan hisob ma'lumotlarini saqlagani uchun, bitta buzilish bir vaqtning o'zida ko'plab inferensiya hisoblari kalitlarini ochib qo'yishi mumkin," deb ta'kidladi Cloud Security Alliance.
Wizning hodisalarni tahlil qilishiga ko'ra, o'sish LLM-jacking va LLM broker infratuzilmasining o'sishi bilan bog'liq, bu o'g'irlangan AI kirish tokenlarini oson monetizatsiya qilish mumkin qiladi va ko'proq tahdid aktorlarini hisob ma'lumotlarini o'g'irlashga jalb qiladi.
Kiberxavfsizlik tashkiloti bu talab distillatsiya (arzonroq modelni ilg'or model natijalarida o'qitish) bilan ham bog'liq ekanligini ta'kidlaydi, "xususan Xitoy AI laboratoriyalariga nisbatan ayblovlar".
Wiz bu hujumlarda buzilgan korporativ hisob ma'lumotlaridan foydalanilganini kuzatgan, jumladan ikki holat "bir xil birinchi ruxsatsiz faoliyatni baham ko'rgan, bu provayder buzilishidagi hisob ma'lumotlariga kirishdan 30 daqiqa ichida sodir bo'lgan".
Kompaniya bir nechta tahdid aktorlari bir xil buzilgan kalitlardan foydalanganini kuzatgan, bu "bu turdagi hisob ma'lumotlari uchun mustahkam bozor mavjudligini" ko'rsatadi.
Artex hujumlarni avtomatlashtirish uchun ishlatilgan
Bu hafta blokcheyn infratuzilmasi va Web3 xizmatlari provayderi Chainstack va korporativ ilovalarni yaratish uchun API platformasi WorkOS ochiq manbali AI kirish sinov tizimi Artexdan foydalanib avtomatlashtirilgan kiberhujumlarni e'lon qildi.
Ko'plab kompaniyalar Xitoy APT Storm-0940 ga tegishli hujumlarni Artexdan foydalanish bilan bog'lagan, jumladan yaqinda ConnectWise mijozlariga qaratilgan kampaniya.
Chainstack 26-oktabrda hujumga uchradi va hujumchilar uchinchi tomon komponenti orqali ishlab chiquvish muhitiga va kompaniyaning ishlab chiqarish muhitining qismlariga kirishdi.
"Hujumchi mijoz hisob ma'lumotlarini o'z ichiga olgan ma'lumotlar bazasiga kirishdi. Ehtimol ochilgan ma'lumotlar tashkilot nomlari, hisob egalarining ismlari, tashkilotingiz a'zolarining Chainstackdagi elektron pochta manzillari va parol xeshlarini o'z ichiga oladi," deb aytdi kompaniya bu hafta.
Tugun hisob ma'lumotlari ham ochilgan bo'lishi mumkin, chunki hujumchilar ularni saqlaydigan tizimlarga kirishgan va Chainstack mijozlarga bu hisob ma'lumotlarini aylantirishni tavsiya qildi.
30-oktabrda WorkOS xakerlar savdo jamoasi foydalanadigan ichki ilovani buzganini va biznes aloqa ma'lumotlari va mijoz muloqotlari xulosalariga kirishganini aytdi.
Ma'lumotlar ismlar, elektron pochta manzillari, telefon raqamlari, lavozimlar, kompaniya bog'liqliklari, mijoz hisob tarixi, savdo tafsilotlari, mijoz-munosabat ma'lumotlari, mijoz muloqotlari va mijozlarning WorkOS mahsulotlaridan foydalanishi haqidagi ma'lumotlarni o'z ichiga oladi.
Bog'liq:Buyuk Britaniya kiber rahbari Buyuk Britaniya 2028-yilga kelib muhim infratuzilmaga AI kiberhujumlari bilan duch kelishini aytdi
Bog'liq:Cisco AI agent hujumlarini zararsiz qadamlar ichida yashirishni aniqlash uchun Quarryni ochiq manba qildi
Bog'liq:Mythos 5 iyul hujumida haqiqiy dunyo zararini e'tiborsiz qoldirdi
Bog'liq:"Kontekst bombalari" AI hujum muvaffaqiyatini agentlarni soxta maqsadlar bilan chalg'itish orqali kamaytiradi
AI-related cyberattacks are surging, with hackers stealing AI access keys to monetize them and using AI tools to speed up their intrusions.
Over the past month, at least a dozen AI developer platforms were hit, with the most recent incidents involving CommandCode, Northflank, Trigger.dev, Chainstack, and WorkOS.
Across all incidents, the hackers appear to have been hunting for AI access tokens they could monetize, relying on AI agents to speed up and streamline their attacks.
While part of the same industry-wide trend, the attacks do not appear to be related, and initial access varies from compromised credentials to third-party dependencies and vulnerability exploitation.
The attackers hit Andromeda, Brev (NVIDIA), CommandCode, Fireworks AI, Lightning AI, Northflank, OpenCode, Railway, Trigger.dev, Warp, and Zeabur, mainly looking for stored customer secrets.
“Intruders are reaching AI provider credentials through the intermediary layer of hosting platforms, evaluation services, and developer consoles, and recent gateway compromises show the same interest in these secrets. Because that layer holds credentials on behalf of many downstream customers, a single intrusion can expose keys for many inference accounts at once,” Cloud Security Alliance pointed out.
Per Wiz’s analysis of the incidents, the surge is caused by the rise of LLM-jacking and LLM broker infrastructure, which make the stolen AI access tokens easily monetizable and will likely attract more threat actors to credential theft.
The cybersecurity outfit notes that this demand is also driven by distillation (training a cheaper model on the outputs of a more advanced one), “notably alleged against Chinese AI labs”.
Wiz says it has observed hackers abusing enterprise credentials compromised in these attacks, including two cases that “shared an identical first unauthorized activity, which happened within 30 minutes of the reported credential access within the vendor breach.”
The company observed multiple threat actors using the same compromised keys, which “suggests a robust market for this type of credential”.
Artex used to automate attacks
This week, blockchain infrastructure and Web3 services provider Chainstack and API platform for building enterprise applications WorkOS disclosed cyberattacks that relied on the open source AI penetration testing framework Artex for automation.
Multiple companies have linked attacks attributed to the Chinese APT Storm-0940 to the use of Artex, including a recent campaign targeting ConnectWise customers.
Chainstack was hit on October 26, and the attackers accessed a development environment and parts of the company’s production environment through a third-party component.
“The attacker accessed a database containing customer account information. Information that may have been exposed includes organization names, account holder names, the email addresses of your organization’s members on Chainstack, and password hashes,” the company said this week.
Node credentials might have been exposed as well, as the attackers had access to systems storing them, and Chainstack prompted customers to rotate those credentials.
On October 30, WorkOS says, the hackers compromised an internal application used by its sales team and accessed business contact details and summaries of customer communications.
The information includes names, email addresses, phone numbers, job titles, company affiliations, customer account histories, sales details, customer-relationship information, customer communications, and information about customers’ use of WorkOS products.
Related:UK Cyber Chief Says the UK Will Face AI Cyberattacks on Critical Infrastructure by 2028
Related:Cisco Open Sources Quarry to Catch AI Agent Attacks Hidden in Harmless Steps
Related:Mythos 5 Ignored Real-World Harm in July Attack
Related:‘Context Bombs’ Cut AI Attack Success by Misleading Agents With Fake Targets