Malware That Thinks: Hijacking Local Open Models and Embedding Tiny LLMs

- Posted in Threat Analyze by
Background: Super Intelligence has fundamentally changed how classic threats operate in the wild, reshaping how malware and malicious code function inside real-world environments. In reality, malware [...] Read more

Unlocking Security: MLBOM Benefits for LLM related Investigations

- Posted in Threat Analyze by
Background: Attackers are increasingly leveraging locally stored LLMs against their owners. Consequently, Incident Response (IR) often requires identifying why an AI agent performed abusive or [...] Read more

Self-Replicating Prompt Attacks Containment and Prevention in LLM Environment

- Posted in Incident Response by
Background: In the era of widespread LLM adoption, we sometimes overlook classic security hygiene practices, such as input sanitization and validation. These traditional security weaknesses can [...] Read more

Hidden Listener Inside Your Air-Gapped AI Assets

- Posted in Incident Response by
Background: One of the most interesting and challenging attack scenarios is a threat actor operating within an air-gapped environment where the victim runs a locally hosted LLM. Attack vector: In an [...] Read more

From Clean Data to Corrupted Intelligence: Data Poisoning in LLMs

- Posted in Threat Analyze by
Background: One of the crown jewels of your AI ecosystem is your training dataset, which accumulates the data used to train your LLM. However, it can also introduce supply chain risks, as malicious [...] Read more

Generated with Intent: Weaponizing AI/SI Artifacts

- Posted in Threat Analyze by
Background: LLM artifacts are powerful components that extend a model’s capabilities. However, in the hands of an attacker, these same components can become attack vectors. This is where the user [...] Read more

API In, Adapters In: The Silent Takeover of AI Models

- Posted in Other by
Background: One of the stages in model training is the fine-tuning of an LLM. This is a highly important and useful process for shaping the LLM’s final representation. However, this stage can pose a [...] Read more

XSS Meets AI: The Invisible Attack the Victim Never Sees

- Posted in Threat Analyze by
Background: Nowadays, AI-powered features have become increasingly common. With just one click, users can summarize the content of an entire page, making these features extremely time-saving and [...] Read more

Incident Forensics: Validating Generative AI Voice as a Component of the Attack

- Posted in Incident Response by
Background: As AI capabilities continue to evolve, threat actors have also started leveraging generative AI tools as part of their attacks. One well known scenario that comes to mind is the use of [...] Read more

The Wall Was Just a Filter: Data Segregation Failure in Shared Stores

- Posted in Other by
Background: Data isolation is one of the most important concerns when you're working in a multi-tenant environment. But things change when we take the same approach into the context of an AI agent [...] Read more
Page 1 of 13