Malware groups develop techniques to evade AI analysis
Cisco Talos reports that several malware families are embedding natural-language instructions in code to confuse or bypass AI analysis tools. These so-called 'AI-analysis evasion' techniques interfere with automated threat detection, with attackers aiming to manipulate language models during triage and classification. Cisco identified this tactic in four malware families over the past 18 months. The best techniques succeeded about 35% of the time, according to their testing.
- AI-evading techniques detected in four malware families
- Techniques embed instructions to confuse AI analysis
- Success rate of evasion is about 35% in testing
- Researchers classify these as 'A3: AI-Analysis Evasion'
- Methods found in both commercial and open-source security environments
Sources covering this
More in Cybersecurity
US and allies disrupt Chinese hacking tools targeting infrastructure
US authorities and international partners have seized domains and tools used by China's Integrity Technology Group to support…
AI-powered attacks target South Korean banks
A cyberattack campaign targeting South Korean financial institutions used AI-based tools to steal personal data, CrowdStrike reported.
Google promotes passkeys as a password alternative
Google is highlighting passkeys as a faster and safer alternative to traditional passwords for its accounts.
Malware found pre-installed on low-cost Android phones
Researchers have discovered malware embedded in the firmware of inexpensive Android smartphones, allowing attackers to remotely install…