Cognitive stress, also known as mental workload, constitutes a central topic within the field of psychophysiology due to its role in modulating attention, autonomic regulation, and stress reactivity. Furthermore, it bears direct relevance to practical monitoring systems that employ non-invasive sensing techniques. This study investigates whether a multimodal, non-invasive measurement setup can detect systematic physiological differences between Resting periods and short episodes of cognitive load within the same individuals. Additionally, it explores the capacity of such a system to differentiate tasks characterized by varying cognitive demands. A sequential, within-subject protocol was employed, comprising five consecutive phases (rest 1, Stroop, rest 12, subtraction, rest 3), during which five modalities were recorded concurrently: EEG, heart rate (HR), galvanic skin response (GSR), facial surface temperature, and oxygen saturation (SpO₂). Beyond phase-wise inspection of time-series data, an exploratory assessment of similarity across participants was conducted using correlation coefficients. The maximum cross-participant correlations observed were 0.88 (HR), 0.90 (GSR), 0.83 (facial temperature), and 0.77 (SpO₂); however, these correlations were used only as exploratory descriptors of inter-individual similarity and did not imply a significant phase effect. For inferential analysis, phase-wise epoch means were evaluated through one-factor repeated-measures ANOVA. The heart rate exhibited a robust main effect of phase (F(4, 32) = 10.5862, p_GG = 0.01044, ηp² = 0.5696), with higher HR observed during cognitive load epochs (e.g., 77.841 ± 11.777 bpm at rest 1 versus 83.926 ± 14.532 bpm during subtraction). The relatively large standard deviation reflects variability between subjects rather than variability within epochs. Regarding processed baseline-referenced GSR, the omnibus phase effect was not statistically significant under the conservative Greenhouse–Geisser correction; therefore, GSR was interpreted as exploratory in this dataset. Facial temperature and SpO₂ likewise did not show statistically significant omnibus phase effects under Greenhouse–Geisser correction (e.g., SpO₂: p_GG = 0.1209). EEG-derived measures provide supplementary central evidence of task engagement; entropy variations within an approximate dynamic range of 0.2 to 0.8 were observed, and the α/θ ratios demonstrated nearly a twofold distinction between rest and cognitive load epochs across different leads.

In recent years, malware and particularly ransomware has evolved rapidly, em-ploying increasingly sophisticated techniques to infiltrate and compromise mod-ern computing environments. This escalation highlights the urgent necessity for adaptive endpoint defense mechanisms capable of detecting and mitigating emerg-ing threats in real time. The present work introduces a Machine Learning-based Endpoint Detection and Response (ML EDR) model designed to provide dynam-ic, behavior-driven protection against both malware and ransomware attacks. The proposed system leverages multi-layer telemetry collected from distributed end-point sensors used to model system behavior, identify anomalies, and predict ma-licious actions before they fully execute. By analyzing traffic flows, process ac-tivity, and file operations in a dynamic sandboxed environment, the model learns behavioral signatures directly from infected samples. These signatures enable the system to classify threats and anticipate future malicious steps with high accura-cy. Integration with a Cloud Intelligence Network and cloud-assisted file analysis enhances the model’s adaptability, enabling rapid updates, collaborative threat in-telligence sharing, and large-scale pattern correlation. The multi-layer monitoring architecture ensures continuous visibility across endpoints, enabling early-stage detection of polymorphic and zero-day ransomware variants. Experimental results demonstrate that the adaptive ML-based EDR model improves detection precision and significantly reduces response time to emerging threats. The study contrib-utes to a scalable, self-evolving defense mechanism suitable for modern enter-prise security ecosystems.

Nowadays smart digital reality is already evolving to a new fundamental change, related to the fast AI algorithmic sentient progress with vast interconnectivities and the resulting H-M symbiotic development. The study outlines a practical experiment on the “digital cognition” modelling, in the context of AI sentient evolu-tion exploration. Expert data and generative AI feedbacks are initially used in this process, both fused in an ad hoc created system-of-systems dynamic model. Fur-ther, a dynamic assessment of the model is performed with mixed H-M validation by implementing the scenario method and probabilistic HPC simulations. Most prominent results have been finally jointly verified in an interactive simulation with a human-in-the-loop active role during CYREX 2025 and in lab conditions. The approach is trying to identify key synergies and gaps of the model, outlining also internal triggers, aiming potential system advancements. The obtained results are expected to support the development of a forward-thinking holistic proactive analysis of the cyber security evolution, giving accent to the sentient features of “digital cognition” in both offensive and defensive context. These will create a suitable post-information society preparation for successful handling of the future interconnected reality’s AI-influenced transcendents and the necessity for a dual human-machine smart regulations establishment within the not so far future.

The rapid advancement of Artificial Intelligence and Large Language Models (LLMs) presents new opportunities for analyzing large-scale unstructured textual data, such as national security strategies. However, effective, machine-assisted methodologies for the qualitative analysis of such documents still remain under-developed. This study proposes opportunities for application of a hybrid method-ology that integrates computational text analysis with the sociological-linguistic framework of “graphing the graphs” by John W. Mohr and the dramatistic pentad of “grammar of motives” by Kenneth Burke. Applying natural language pro-cessing (NLP) techniques, including: named entity recognition (NER), sentiment analysis via specialized dictionaries, and contextual relationship extraction using tools like Voyant Tools – the framework enables the machine-assisted identifica-tion of key Actors, Actions, and Scenes within strategic texts. Experimentally ap-plied to three Bulgarian national security documents (1998–2018), the approach demonstrates the extraction of dominant terminology, key actors (Bulgaria, NATO, EU), and their contextual relationships. The results validate the potential of AI to support strategic analysis while highlighting limitations such as model hallucinations and data sensitivity risks. The proposed setup explicitly calls for further integration of machine learning, including advanced models for relation extraction and hybrid validation systems, aiming to increase automation, accura-cy, and scalability in the analysis of strategic narratives.

This study proposes a machine learning and deep learning framework to detect application layer DDoS attacks using web server access logs. A realistic dataset was generated using eight DDoS tools, along with benign traffic. Multiple ML models, including Random Forest, SVM, and KNN, as well as DL models such as ANN and LSTM, were evaluated. The results show high detection performance, with the LSTM achieving up to 99% accuracy and outperforming other models. Additional validation on non-DDoS attacks revealed that ML models performed poorly, while LSTM maintained strong performance (92.89%). The findings highlight the effectiveness of log-based datasets and the superiority of LSTM for handling sequential and mixed-feature data in cybersecurity detection tasks.