As AI tools become more integrated into daily work, the security risks attached to them are also evolving in new directions. Researchers at Trail of Bits have demonstrated a method where malicious prompts are hidden inside images and then revealed during processing by large language models.
The technique takes advantage of how AI platforms downscale images for efficiency, exposing patterns that are invisible in their original form but legible to the algorithm once resized.
Read more…
Source: TechPro News
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- RondoDox botnet linked to large-scale exploit of critical HPE OneView bug
January 16, 2026
A critical HPE OneView flaw is now being exploited at scale, with Check Point tying mass, automated attacks to the RondoDox botnet. The security outfit says it has identified “large-scale exploitation” of CVE-2025-37164, a maximum-severity remote code execution bug in HPE’s data center management platform. Check Point has tied the activity to RondoDox, a Linux-based botnet ...
- WhisperPair exposes Bluetooth earbuds and headphones to tracking and eavesdropping
January 16, 2026
WhisperPair is a set of attacks that lets an attacker hijack many popular Bluetooth audio accessories that use Google Fast Pair and, in some cases, even track their location via Google’s Find Hub network—all without requiring any user interaction. Researchers at the Belgian University of Leuven revealed a collection of vulnerabilities they found in audio accessories ...
- German police add Black Basta boss to EU most-wanted list
January 16, 2026
German cops have added Russian national Oleg Evgenievich Nefekov to their list of most-wanted criminals for his services to ransomware. Nefekov, 35, is accused of spearheading the Black Basta ransomware operation, which suffered a similar fate as Conti last year – ceasing activity after a major internal leak. His name and face also now appear ...
- Cisco has finally patched a maximum-level security issue
January 16, 2026
A maximum-severity vulnerability in certain Cisco products has finally been addressed after allegedly being exploited by Chinese hackers for several weeks. In mid-December 2025, the networking giant disclosed a remote code execution (RCE) vulnerability in AsyncOS that affects Secure Email Gateway (SEG) and Secure Email and Web Manager (SEWM) appliances. It tracked the flaw as CVE-2025-20393 ...
- “Reprompt” attack lets attackers steal data from Microsoft Copilot
January 15, 2026
Researchers found a method to steal data which bypasses Microsoft Copilot’s built-in safety mechanisms. The attack flow, called Reprompt, abuses how Microsoft Copilot handled URL parameters in order to hijack a user’s existing Copilot Personal session. Copilot is an AI assistant which connects to a personal account and is integrated into Windows, the Edge browser, and ...
- DeadLock Ransomware: Smart Contracts for Malicious Purposes
January 15, 2026
DeadLock is a ransomware family discovered in July 2025. It is notable for not being associated with any known affiliate programs and for lacking a Data Leak Site (DLS). This, combined with the limited number of reported victims, has resulted in low exposure for the group. However, Group-IB specialists have discovered an interesting use of ...