[Photo: Reve AI]
[DigitalToday reporter Chi-gyu Hwang (황치규)] AI hacking techniques are surging in volume, and a shift toward higher sophistication is becoming pronounced.
AI-based hacking is still in an early stage, but an analysis says a growing number of free open-source tools is fundamentally changing the economics of cybercrime. Google has for the first time confirmed a case in which attackers fully automated vulnerability discovery and attacks using an LLM framework without human involvement. Google expects that within the next three months, simple attacks carried out by multiple AI agents will emerge with little human involvement. It also forecasts there is a strong possibility that sophisticated, fully autonomous attacks will appear within six months.
ㆍOpen-source AI hacking tools spread, changing the economics of cybercrime
ㆍ”Possibility of fully autonomous AI cyberattacks emerging within six months” – Google report
Moves by the industry toward security-focused AI models are also accelerating. Microsoft introduced a preview of its security-focused model, “MAI-Cyber-2-Flash”. Security-focused AI model startup Corma raised $125 million in Series A funding. The company said existing AI models are better at attacking than blocking attacks, making defense-focused models necessary. Palo Alto Networks will acquire AI model training and inference startup Applied Compute for $4.5 billion. Applied Compute was founded in 2025 by former OpenAI researchers and helps companies further train and operate open-weight models.
ㆍMicrosoft unveils AI agent “self-improvement” loop, also expands its own AI model lineup
ㆍWhy security-focused models are needed? “Existing AI models have stronger attack capabilities than defense”
ㆍPalo Alto Networks to acquire AI model training and inference startup Applied Compute for $4.5 billion
It also summed up moves and issues around security involving companies in and outside South Korea.
OpenAI unveiled two security products that use AI agents to find software vulnerabilities and produce fixes. Anthropic will cut the maximum disclosure grace period for software vulnerabilities to 30 days from 90 days as AI’s ability to detect security flaws rapidly improves. It decided to review existing disclosure practices in the security industry as the likelihood increases that AI will repeatedly discover the same vulnerabilities.
ㆍOpenAI unveils two security services, automating vulnerability discovery and patching
ㆍ”AI finds vulnerabilities too fast”…Anthropic shortens disclosure deadline from 90 days to 30 days
F5 presented three key changes in its major cybersecurity outlook for 2027: securing security sovereignty as sovereign AI spreads, adopting multimodel AI for operational stability, and accelerating AI-based cyberattacks. Check Point introduced its data center security solution, “Maestro Hyperscale Orchestrator (MHO) 200”. An incident in which a malicious “Claude Code” extension was distributed on the Cursor marketplace for an AI coding platform reportedly began with access keys leaked in a Fireworks AI hack.
ㆍ”In the 2027 security market, watch security sovereignty, multimodel use and accelerated AI hacking”
ㆍCheck Point launches Maestro MHO 200, strengthens AI data center security
ㆍCursor malicious Claude Code extension incident traced to Fireworks AI hack
ㆍAI development platform Fireworks AI hacked again, some customer data including Notion leaked
AI-based cybersecurity company Tiori found five critical vulnerabilities in the web server software Nginx that allow unauthenticated remote code execution (RCE). Offensive security specialist Enki Whitehat completed a transition audit for ISO 27001, an international standard for information security management systems, and obtained ISO 27017 certification, an international standard for cloud service security. S2W Lab took part in the AI Expo Tokyo 2026 held from the 11th to the 13th at Makuhari Messe in Chiba prefecture, Japan, and showcased an AI agent-based integrated security platform.
ㆍTiori finds critical Nginx vulnerabilities, recommends patching
ㆍEnki Whitehat obtains international information security and cloud security standard certifications
ㆍS2W Lab showcases AI-based integrated security platform at AI Expo Tokyo 2026
AISpera launched an AI-based threat exposure management platform, “Criminal IP AITEM”. RaonSecure will integrate Google’s “Device Trust from Android Enterprise” with its authentication and device management solutions and will participate as an Android Enterprise partner. FASOO AI held an event for people responsible for security and AI transformation (AX) and shared strategies to achieve both maximising AI value and security.
ㆍAISpera launches AI-based threat exposure management platform AITEM
ㆍRaonSecure participates as a Google Android Enterprise partner
ㆍFASOO AI CEO Gyu-gon Cho (조규곤): “In AX strategy, it is also important to decide whether to build AI agents directly or buy and use them”
South Korea’s National Intelligence Service, together with the National Security Research Institute, published standards for cryptographic safety for the state and public institutions and a guidebook for transitioning to post-quantum cryptography on the National Cyber Security Center website.
ㆍNIS releases public-institution cryptographic safety standards and post-quantum cryptography transition guidebook
Click Here For The Original Source
——————————————————–
