Plainly speaking, Artificial intelligence is no longer a fringe technology. It has become a core component of modern business, from customer service chatbots to complex data analysis.
We often treat the Large Language Models (LLMs) that are at the core of this technology as trusted black boxes. But like any software, they can be tampered with, manipulated, and turned against their creators. Understanding the ways an AI model can be compromised is the first step toward building a secure and resilient AI infrastructure. There are three primary ways that the authors want to cover, in which they will explain how an AI model can be compromised and made to act in unpredictable (or malicious!) ways:
Read more…
Source: Trend Micro
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Researcher uncovers dozens of sketchy Chrome extensions with 4 million installs
April 11, 2025
Google is hosting dozens of extensions in its Chrome Web Store that perform suspicious actions on the more than 4 million devices that have installed them and that their developers have taken pains to carefully conceal. The extensions, which so far number at least 35, use the same code patterns, connect to some of the same ...
- GOFFEE continues to attack organizations in Russia
April 10, 2025
GOFFEE is a threat actor that first came to our attention in early 2022. Since then, Kaspersky researchers have observed malicious activities targeting exclusively entities located in the Russian Federation, leveraging spear phishing emails with a malicious attachment. Starting in May 2022 and up until summer of 2023, GOFFEE deployed modified Owowa (malicious IIS module) in ...
- Incomplete NVIDIA Patch to CVE-2024-0132 Exposes AI Infrastructure and Data to Critical Risks
April 10, 2025
In September 2024, NVIDIA released several updates to address a critical vulnerability (CVE-2024-0132) in its NVIDIA Container Toolkit. If exploited, this vulnerability could expose AI infrastructure, data, or sensitive information. With a CVSS v3.1 rating of 9.0, all customers were advised to update their affected software immediately. Further research, however, uncovered that the patch was incomplete. ...
- Password Spray Attacks Taking Advantage of Lax MFA
April 10, 2025
In the first quarter of 2025, Rapid7’s Managed Threat Hunting team observed a significant volume of brute-force password attempts leveraging FastHTTP, a high-performance HTTP server and client library for Go, to automate unauthorized logins via HTTP requests. This rapid volume of credential spraying was primarily designed to discover and compromise accounts not properly secured by multi-factor ...
- Operation Endgame follow-up leads to five detentions and interrogations as well as server takedowns
April 9, 2025
Following the massive botnet takedown codenamed Operation Endgame in May 2024, which shut down the biggest malware droppers, including IcedID, SystemBC, Pikabot, Smokeloader and Bumblebee, law enforcement agencies across North America and Europe dealt another blow to the malware ecosystem in early 2025. In a coordinated series of actions, customers of the Smokeloader pay-per-install botnet, operated ...
- Court document reveals locations of WhatsApp victims targeted by NSO spyware
April 9, 2025
NSO Group’s notorious spyware Pegasus was used to target 1,223 WhatsApp users in 51 different countries during a 2019 hacking campaign, according to a new court document. The document was published on Friday as part of the lawsuit that Meta-owned WhatsApp filed against NSO Group in 2019, accusing the surveillance tech maker of exploiting a ...