A new virus with AI-powered 🚨

Google Threat Intelligence Group, or GTIG, has revealed a new report that talks about major changes in cyberspace because hackers now no longer use AI, just to help them work or write code, but to put AI in malware to "think and adapt themselves" during real attacks. It's called a new era of cyber warfare where attack tools can learn and evade detection on their own.

.

Newer malware found by Google, such as PROMPTFLUX and PROMPTSTEAL, uses a large language model or LLM during work to reconstruct dangerous scripts every time PROMPTFLUX runs. It is written in VBScript and can send commands to Gemini's API to help rewrite complex and self-encrypted code to dodge antiviruses. PROMPTSTEAL was used by the Russian APT28 group in an attack on Ukraine. It fakes an image and uses the Qwen model to create commands that steal local data without pre-written code.

.

Interestingly, some hackers are starting to use Social Engineering and AI techniques to trick models into writing dangerous code, using seemingly experimental or educational excuses, such as pretending to be a Capture-the-Flag contestant to get Gemini to introduce vulnerabilities, or quoting as a student completing a project to ask AI to write code. All this reflects that hackers are no longer just fooling people, but are also fooling AI.

.

The report also says that the black market of AI tools for attack is also growing dramatically by 2025. There are tool sales services like WormGPT, FraudGPT and LoopGPT that help do everything from write phishing emails, create malware, to find system vulnerabilities.

.

This makes it even easier for novice hackers to launch more sophisticated attacks. Meanwhile, state-sponsored groups such as North Korea, China and Iran use these AI at every stage of their attacks, from finding information, preparing fisching, to developing control servers and stealing data.

.

To counter this situation, Google has closed accounts and projects related to the ill-fated, improving the Gemini model and preventing abuse, as well as collaborating with the DeepMind team to develop tools like Big Sleep and CodeMender that use AI to automatically detect and repair vulnerabilities. The goal is to create an AI that is both advanced and secure so that humans can use technology responsibly in an era when AI is both a weapon and an armor at the same time.

.

Source: Google

# IT News # Includes IT matters # Cough to know # IT