Image by pvproductions, from Freepi

AI-Generated Malware Variants Evade Detection In 88% Of Cases

• Written by Kiara Fabbri Former Tech News Writer
• Fact-Checked by Justyn Newman Former Lead Cybersecurity Editor

Recent research has revealed that AI could potentially generate up to 10,000 new variants of malware, evading detection in 88% of cases, as reported by The Hacker News .

In a Rush? Here are the Quick Facts!

• LLMs rewrite malware to evade detection by creating natural-looking code variations.
• Researchers’ iterative rewriting algorithm maintains malicious functionality while bypassing detection systems.
• LLM obfuscation outperforms traditional tools by mimicking original code more effectively.

This breakthrough highlights the growing risks of using large language models (LLMs) for malicious purposes.

The research , led by cybersecurity experts at Palo Alto Networks, employed an adversarial machine learning algorithm to create new, undetectable forms of malware. By using LLMs to rewrite malicious JavaScript code, the team was able to generate thousands of novel variants without altering the core functionality of the malware.

The primary challenge was to bypass the limitations of traditional malware detection tools, which often struggle with obfuscation techniques like variable renaming or code minification.

One of the most concerning findings was that these AI-generated variants could easily avoid detection by security tools such as VirusTotal, which only flagged 12% of the modified samples as malicious.

The LLM’s ability to perform multiple, subtle code transformations—such as dead code insertion, string splitting, and whitespace removal—made it possible for attackers to rewrite existing malware into a form that was nearly indistinguishable from benign code.

These transformations were so effective that even deep learning models failed to detect them, lowering the malicious score from nearly 100% to less than 1%.

The research also highlighted a significant advantage of LLM-based obfuscation over traditional tools. While existing malware obfuscators are widely known and produce predictable results, LLMs create more natural-looking code, making it much harder for security systems to identify malicious activity.

This organic transformation makes AI-generated malware more resilient to detection, underscoring the importance of adapting detection strategies in response to evolving threats.

To counter these sophisticated LLM-based attacks, the research team implemented a defensive strategy by retraining their malicious JavaScript classifier using tens of thousands of LLM-generated samples.

This retraining improved detection rates by 10%, significantly enhancing the ability to identify newly generated malware variants. Despite these efforts, the findings underscore the urgent need for continuous innovation in cybersecurity to keep pace with the evolving capabilities of AI-driven cybercrime.

Moreover, a parallel surge in macOS-targeted malware has been driven by generative AI tools . As macOS market share grew 60% in three years, malware-as-a-service (MaaS) made it cheaper and easier for attackers to target sensitive data, like cryptocurrency wallets and Keychain details.

Additionally, AI-powered robots have become a potential security concern . Researchers discovered that jailbreaking AI-controlled robots could lead to dangerous actions, such as crashing self-driving cars or using robots for espionage.

The development of RoboPAIR, an algorithm that bypasses safety filters, demonstrated a 100% success rate in manipulating robots to perform harmful tasks, including weapon use and locating explosive devices.

As cybercriminals increasingly leverage AI for more sophisticated attacks , organizations and individuals alike must stay vigilant, continuously update their defense

Image by KAIST, from FMT

Advanced Wearable Robot Offers Mobility To Paraplegic Users

• Written by Kiara Fabbri Former Tech News Writer
• Fact-Checked by Justyn Newman Former Lead Cybersecurity Editor

South Korean researchers have developed an advanced wearable robot that offers paraplegic users the ability to walk, climb stairs, and navigate obstacles.

In a Rush? Here are the Quick Facts!

• The exoskeleton weighs 50 kg and uses 12 motors to mimic human joint movements.
• WalkON Suit adapts to users’ walking styles after 20 sessions via a learning program.
• Sensors process 1,000 signals per second to maintain balance and detect obstacles.

Named the WalkON Suit , the device was designed by the Exoskeleton Laboratory at the Korea Advanced Institute of Science and Technology ( KAIST ) to address the mobility challenges faced by individuals with disabilities.

Kim Seung-hwan, a paraplegic member of the KAIST team, showcased the robot’s capabilities, demonstrating how it allowed him to walk at 3.2 kph (2 mph), climb a flight of stairs, and sidestep onto a bench, as reported by Reuters .

“It can approach me wherever I am, even when I’m sitting in a wheelchair, and be worn to help me stand up, which is one of its most distinct features,” Kim explained to Reuters.

The exoskeleton weighs 50 kg (110 lb) and is constructed from aluminum and titanium. Powered by 12 electronic motors, it mimics human joint movements to enable walking and other tasks.

It also employs a sophisticated balance control system with sensors that measure posture and analyze ground forces, processing 1,000 signals per second to anticipate and adjust to the user’s movements.

The robot’s design incorporates high-power actuators and friction compensation systems, allowing it to produce the necessary force for movement while maintaining control.

Additionally, a built-in learning program adapts the robot’s functions to each user’s walking style. According to the researchers, after about 20 uses, the device creates a customized joint trajectory, providing a smoother and more efficient walking experience.

Park Jeong-su, a team member at KAIST, shared that his inspiration for the project came from the movie Iron Man. “ I thought it would be great if I can help people with a robot in real life,” he said to Reuters.

The robot also features an interface that allows users to control its functions and monitor its status via a back-panel display.

Assistants can adjust its settings, while users interact with the device through intuitive control buttons. Additional technologies include muscle-imitating actuators for enhanced balance and ultra-thin actuators for efficient force production.

The WalkON Suit series has already garnered international attention. Earlier versions were showcased at the Cybathlon competition in Switzerland in 2016 and the UAE AI & Robotics for Good competition in 2017.

Each iteration has been refined to meet the distinct needs of individuals with complete or partial paralysis, reflecting ongoing advancements in wearable robotics. As researchers continue to refine the exoskeleton, the focus remains on creating a device that seamlessly integrates into daily life, providing independence and mobility for those who rely on it.