BotNet News

Cyberthreat News is a daily digest of the latest cybersecurity threats, data breaches, and hacks that affect your business. Read articles, videos and podcasts from the experts in information security to learn how to protect your organization.

A new jailbreak technique could help adversaries bypass large language models (LLM) safety guardrails and produce potentially harmful or malicious responses. The multi-turn attack, codenamed Bad Likert Judge by Palo Alto Networks Unit 42 researchers Yongzhe Huang, Yang Ji, Wenjun Hu, Jay Chen and Akshata Rao, essentially asks the LLM to act as a “judge” scoring the harmfulness of a response using a set of ratings scales.

Threat actors use malware and phishing attacks to steal sensitive data, including personally identifiable information (PII), credit card details and passwords. They can also launch denial-of-service attacks that prevent computers from accessing critical resources and services.

Terrorist groups and cybercriminals are less developed in their use of attacks and have a lower propensity to pursue cyber means than nation-states, but they still pose significant risks. Cross-domain attacks, a tactic used by adversaries to exploit weaknesses across multiple domains, are a major concern as they enable attackers to move laterally between environments, avoid detection and evade mitigations.