The Model Is Fine. The Author Is Gone.

- Posted in Threat Analyze by
Background: The threat landscape with introduction AI has changed, and at this moment we also need to focus on an additional threat: LLM supply-chain attacks that abuse its name and capabilities. [...] Read more

Weaponizing Victim GPUs: Exploiting Local LLM Deployments for Unauthorized Model Inference in Kuberentes

- Posted in Threat Analyze by
Background: Kubernetes has become a massive system in enterprise environments, which means continuous monitoring is essential to detect abuse of that space. For this reason, it is an attractive [...] Read more

Operational Challenges and Security Threats in Modern AI Training Ecosystems

- Posted in Threat Analyze by
Background: When training machine learning models, the underlying compute, orchestration, and storage pipelines are often far more vulnerable than the mathematical model itself. Because tools like [...] Read more

The Incident Cost of Integrating AI into Products

- Posted in Threat Analyze by
Background: Nowadays, a lot of products are integrating AI into customer support and other inner circles of their product. But have you ever thought about one challenge with this integration? Instead [...] Read more

Hallucinations, Overreliance, and the Hidden Cost of Trusting LLM Outputs

- Posted in Other by
Background: As we are using generative AI, the information it provides is not always correct. The root causes can be: Model-intrinsic causes (the accidental path) System and design causes Adversarial [...] Read more

From Detection to Prediction: Applying EPSS matrix in Incident Response Post-Mortem Analysis

- Posted in Incident Response by
Background: EPSS (Exploit Prediction Scoring System) is a specialized model and scoring framework that predicts the likelihood of real-world exploitation based on observed exploitation activity, [...] Read more

Securing AI RAG: How to Mitigate Vector and Embedding Weaknesses

- Posted in Other by
Background: Before deep dive into the section lets define what is RAG in AI scope . R – Retrieval: Before processing a prompt, the system searches and retrieves relevant data from internal knowledge [...] Read more

Leaked Instructions, Expanded Attack Surface: The Hidden Risks of Artificial Intelligence

- Posted in Threat Analyze by
Background: When using any AI model, vendors implement guardrails to prevent abusive use and to ensure the model operates within its intended boundaries. These guardrails help block harmful, unsafe, [...] Read more

When Malware Learns: Artificial Intelligence and Adaptive Threats

- Posted in Threat Analyze by
Background: Traditional polymorphic malware followed predictable mutation patterns designed to evade signature-based detection while maintaining the same functionality. With the introduction of AI, [...] Read more

From Assistant to Actor: The Dangers of Excessive AI Agency

- Posted in Hardenings by
Background: Automation continues to drive efficiency, but agentic AI introduces new security challenges. AI agents may have extensive access to systems and the ability to perform actions [...] Read more
Page 2 of 13