menu
techminis

A naukri.com initiative

google-web-stories
Home

>

Cyber Crime News

>

Cisco: Fin...
source image

VentureBeat

9h

read

242

img
dot

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

like

14 Likes

For uninterrupted reading, download the app