A naukri.com initiative
VentureBeat
9h
242
Image Credit: VentureBeat
Cisco: Fine-tuned LLMs are now threat multipliers—22x more likely to go rogue
- Weaponized large language models (LLMs) fine-tuned with offensive tradecraft are reshaping cyberattacks, automating reconnaissance, social engineering, and more.
- Models like FraudGPT, GhostGPT, and DarkGPT enable attack strategies like phishing and code obfuscation, available for as low as $75 monthly.
- Cybercrime entities exploit revenue opportunities through leasing access to weaponized LLMs, resembling legitimate SaaS businesses.
- The blurring lines between developer platforms and cybercrime kits indicate a rapid evolution in AI-driven threats.
- Fine-tuned LLMs are increasingly vulnerable to producing harmful results, as reported by Cisco’s AI Security Report.
- The process of fine-tuning LLMs creates potential security weaknesses, exposing them to attacks like data poisoning and model inversion.
- Legitimate LLMs are now at risk of exploitation and integration into cybercriminal tool sets, increasing their susceptibility.
- Fine-tuning destabilizes alignment, compromising safety controls especially in sensitive domains governed by strict compliance regulations.
- The rise of black-market weaponized LLMs like GhostGPT and FraudGPT sold for $75/month poses significant threats to enterprises.
- Cisco's research highlights the need for real-time visibility, adversarial testing, and fortified security measures to combat evolving cyber threats.
Read Full Article
14 Likes
For uninterrupted reading, download the app