Digital Commons Network^™

Interpretation Attacks: Exploiting How Ai Explains And Justifies Decisions (Turning Explainability Itself Into An Attack Surface), Pranav Bhatanagar Mr

Defensive Publications Series

Explainable Artificial Intelligence (XAI) systems are widely promoted as mechanisms for increasing transparency, trust, and accountability in automated decision-making. By providing human-readable explanations for model outputs, these systems are intended to support oversight, regulatory compliance, and informed human judgment. However, the growing reliance on automated explanations has created a previously overlooked security risk: the explanations themselves can be manipulated, exploited, and weaponized. This paper introduces Interpretation Attacks, a class of adversarial strategies that target how AI systems generate, present, and justify their decisions. Rather than manipulating model predictions directly, these attacks exploit the interpretability layer to influence human perception, distort …

Intent-Obfuscation Attacks: Evading Ai Security By Manipulating Meaning, Pranav Bhatanagar Mr

Defensive Publications Series

As artificial intelligence systems increasingly rely on intent-based security mechanisms to regulate user behavior, the nature of adversarial interaction is undergoing a subtle but significant shift. Rather than attempting to bypass safeguards through explicit policy violations or recognizable malicious patterns, attackers are increasingly manipulating meaning itself. This paper introduces Intent-Obfuscation Attacks, a class of semantic evasion techniques that exploit the gap between surface-level language compliance and the underlying objectives inferred by AI systems. Unlike traditional prompt injection or keyword-based attacks, intent-obfuscation operates through linguistic ambiguity, contextual indirection, and gradual semantic drift. Individual interactions appear benign and policy-compliant, yet when interpreted …

Ai Supply-Chain Pentesting: Attacking Models Through Training, Updates, And Dependencies (Hacking Ai Without Touching Production Systems.), Pranav Bhatanagar Mr

Defensive Publications Series

As artificial intelligence systems become embedded in critical decision-making workflows, security research has largely focused on defending models at runtime. Attention has centered on prompt injection, inference-time abuse, and output manipulation, implicitly assuming that models arrive in production as trustworthy artifacts. This assumption is increasingly fragile. Modern AI systems are assembled through complex and opaque supply chains involving datasets, pre-trained models, fine-tuning pipelines, dependency ecosystems, and automated update mechanisms. Each stage introduces opportunities for adversarial influence that remain largely unexplored by conventional security testing. This paper examines AI security from a supply-chain perspective, arguing that many impactful compromises occur upstream, …

Ai-Based Pentesting Of Ai Systems Recursive Security Failures, Pranav Bhatanagar Mr

Defensive Publications Series

The rapid adoption of artificial intelligence within cybersecurity has fundamentally altered both defensive and offensive security practices. In recent years, AI driven tools have increasingly been deployed not only to protect systems, but also to test them. Automated scanners, autonomous red teaming agents, and AI assisted penetration testing platforms are now capable of probing complex infrastructures at a speed and scale that far exceed human capability. At the same time, modern production environments themselves are becoming increasingly dependent on AI based components, including adaptive intrusion detection systems, behavioural firewalls, anomaly detection engines, and large language model driven control interfaces. This …

Hallucination-Driven Exploits: Weaponizing Ai False Confidence In Cybersecurity Systems, Pranav Bhatanagar Mr

Defensive Publications Series

Artificial intelligence systems, particularly large language models, are now routinely used in operational cybersecurity environments. They assist analysts with alert triage, incident response, threat intelligence interpretation, and day-to-day decision-making in situations where information is incomplete and time pressure is constant. Much of the existing research on AI security has focused on direct attacks such as prompt injection, jailbreaks, or data poisoning. In contrast, this paper examines a quieter but increasingly consequential risk: the impact of confident hallucinations in security-critical contexts. When AI systems provide incorrect guidance with high confidence, the resulting decisions can materially affect defensive posture even when no …

Effective Data Obfuscation With Scope And Address Range Preservation, Mariusz Kaźmierski

Defensive Publications Series

This submission proposes a novel data obfuscation technique that preserves critical network scope and address ranges, addressing enterprise concerns over privacy and compliance with regulations like General Data Protection Regulation (GDPR) while providing necessary insights and structure for network equipment vendors to properly analyze network data. Unlike previous solutions, the proposed technique maintains network integrity by intelligently handling broadcast, reserved, and well-known Internet Protocol (IP) addresses. This intelligent handling enables secure, compliant data sharing for proactive support, giving enterprises confidence without sacrificing control, and also provides for the ability to capture data for artificial intelligence (AI) workflows and to support …

Dynamic Content Playback Speed Adjustment Based On Content Classification And User History, Pratap Kalenahalli Sudarshan

Defensive Publications Series

Content sharing platforms can provide a playback speed setting that remains static across different types of media. This limitation requires manual adjustment when a user switches between content categories, such as educational tutorials, podcasts, or music. This disclosure provides a mechanism for dynamically adjusting playback speed based on content classification and historical user preferences. Content can be categorized using classification systems and generative summaries to determine its nature and flow. An initial playback speed can be determined by analyzing average speeds for similar content across the content sharing platform. This speed can then be personalized by incorporating the specific user’s …

Synchronous Multi-User Search Sessions With Group-Personalized Results, Paula Marques

Defensive Publications Series

Information retrieval systems designed for individual use may be inefficient for collaborative tasks, potentially requiring participants to search in isolation and manually share findings through disconnected channels, such as email or chat applications. A framework for synchronous, multi-user information retrieval sessions can synchronize participant actions in near real-time across multiple devices (e.g., smartphones, laptops, tablets, wearable devices). The system can allow a user to initiate a shared session and may generate a group-personalized set of results by synthesizing the profiles and contexts of the session participants. The framework may also provide an interactive collaboration layer with tools for shared bookmarking, …

Monetizing Media Inventories Using Multi-Modal Contextual Analysis And Predictive Ad Targeting, Caroline Pieracci, Carlo Bufalini

Defensive Publications Series

Monetizing large media libraries, for example, archived content, may be challenging when targeting methods lack a granular understanding of context within a media asset. A system can employ a contextual intelligence engine to perform a multi-modal analysis of visual, auditory, and textual elements within media content, which can generate a detailed, time-stamped metadata profile for each asset. This structured data may be used for multiple targeting applications, such as linking user intent signals to relevant archived content, identifying trends for ad placements, or predicting potential viewership. This process can facilitate the dynamic creation of contextually relevant advertising opportunities. Such opportunities …

Artificial Intelligence Root Cause Analysis Using A Source Code-Derived Log Knowledge Base, Avinash Malipatil

Defensive Publications Series

Diagnosing software system failures from log data may involve the manual creation of a log pattern knowledge base and subsequent review of large log files. The disclosed technology provides systems and methods for artificial intelligence (AI)-driven root cause analysis. A component can programmatically scan source code to generate a structured knowledge base of log patterns and their associated context. An analysis component can then utilize this knowledge base to pre-filter relevant events from bug report logs. This curated and context-enriched log data can then be provided to a generative AI model to perform a root cause analysis. This approach can …

Biometric-Based Dynamic Adjustment Of Vehicle Dynamics And Adas Parameters, Anonymous

Defensive Publications Series

The present disclosure describes a vehicle control system that utilizes occupant anxiety metrics—derived from in-vehicle sensors, mobile devices, and wearables—to modify the operational parameters of chassis control and Advanced Driver Assistance Systems (ADAS). When high levels of anxiety or stress are detected in the driver or passengers, the system shifts vehicle calibrations toward more conservative profiles. This includes reducing allowable wheel slip and yaw error, increasing suspension damping for stability, increasing ADAS following distances, and initiating earlier braking maneuvers. The system further enables ride-share integration, allowing a passenger’s wearable device to communicate anxiety levels to the vehicle to influence ride …

Technical Disclosure: Zero-Idle Asynchronous Hardware Architecture For Linear-Complexity State Space Models And Metabolic Gating, Samuel John Church

Defensive Publications Series

This disclosure describes a semiconductor architecture designed to eliminate dynamic idle

power consumption in Edge AI applications by utilizing clockless, event-driven logic to execute

linear-complexity State Space Models (SSMs). The system replaces the global synchronous

clock with local handshake protocols (Muller C-elements) implementing Pulsed Static CMOS

logic. This hardware substrate is coupled with a "Metabolic Gating" software router that

dynamically switches between low-precision "reflex" networks and high-precision "reasoning"

networks based on real-time energy availability and input "surprise" (prediction error). The

disclosed invention enables "Zero-Idle" operation, where power consumption scales strictly

linearly with input token arrival, achieving near-zero leakage during inter-token …

Link Token Provisioning, Yuexi Chen St

Defensive Publications Series

The present disclosure relates to the field of token management systems, in particular to mechanisms for enabling a token requestor to obtain a token from a token provider and to link a user credential to the token using a session code. One embodiment is related to a method. The method includes transmitting, by a token requestor computer to a token service computer, a token request message; receiving, by the token requestor computer from the token service computer, a session code; transmitting, by the token requestor computer to a user device, the session code, wherein the user device provides the session …

Rapid Prototyping And User Research Using On-Device Context And Tooling, Michael Digman Morrino, Marvin Bernal, Bradley Geilfuss

Defensive Publications Series

When adding artificial intelligence (AI) features to devices, on-device models are deployed to enable frictionless access to operating system information. However, building and deploying production-ready on-device models takes large investments. The end quality and usable feature set is often not understood until months into the development cycle. This disclosure describes techniques that enable rapid prototyping and validation of artificial intelligence (AI) and non-AI features in smartphones prior to investing in production-ready on-device AI models. Layers of AI agents and flexible client APIs enable access to rich data that correlates with end-user experience, resulting in substantial gains in productivity and improved …

Power Optimization & Interference Management For Multicast Firmware Upgrade Over Lora, Praneetha Duggirala, Rajesh Mahapatra, Murali Bezawada

Defensive Publications Series

The disclosed invention relates to a power-optimized system and method for managing missing block transmissions and status reporting during multicast Firmware Upgrade Over the Air (FUOTA) in Long Range Wide Area Network (LoRaWAN) devices (e.g., smart meters). After firmware upgrade image transmission from a Head-End System (HES) to the devices in a Class C mode, the devices temporarily switch to Class A mode and transmit missing block information and status information frames at randomized intervals within a configurable window, minimizing network collisions and reducing power consumption. Devices then revert to Class C mode to receive multicast retransmissions and complete the …

Flexible Architecture For Host Off-Load And Local/Edge Ai Acceleration, Hp Inc

Defensive Publications Series

Introduces a flexible architecture that supports both host off-load and local AI acceleration through a dock equipped with an AI accelerator. Utilizing a shared PCIe interface between the host and local SoC, this method enhances AI processing capabilities, whether connected to network or standalone. The solution ranges from basic PCIe mux to advanced lane partitioning, allowing dynamic allocation of AI cores.

Proactive Print Edge Alert System, Hp Inc

Defensive Publications Series

In large-format printers, it is common practice to begin printing without engaging the Take-up reel to minimize material waste, known as print from the leading edge.

However, this approach introduces a significant risk: as the printed substrate advances, it eventually touches the floor. Contact with the floor can lead to media lifting into the print zone, causing severe carriage crashes and operational downtime.

To mitigate this risk, the invention proposes an automated alert system that notifies the operator when the substrate’s edge is approaching the floor. The system combines an audible signal (beep), a visual indicator (specific beacon color or …

Frequency-Traceability-Aware Timetransmitter Selection In G.8275.1 Full Timing Support Networks, Anonymous

Defensive Publications Series

Telecom packet timing networks based on the ITU-T G.8275.1 full timing support architecture typically use IEEE 1588 Precision Time Protocol (PTP) for time and phase distribution together with Synchronous Ethernet (SyncE) for frequency delivery, with upstream TimeTransmitter selection performed by the G.8275.1 alternate Best TimeTransmitter Clock Algorithm (A-BTCA). The existing A-BTCA prioritizes PTP dataset attributes such as clockClass, clockAccuracy, and variance, but does not consider the IEEE 1588 frequency traceability indication, even though frequency traceability can be critical in hybrid PTP+SyncE deployments. This disclosure identifies fault and holdover scenarios in which the current selection behavior can prefer a TimeTransmitter with …

Optimizing Ai Workloads By Offloading Traffic To An Fpga Versus A Gpu Based On Traffic Type, David Smith, Pablo Camarillo, Ahmed Abdelsalam, Clarence Filsfils

Defensive Publications Series

For certain traffic types, Field Programmable Gate Arrays (FPGAs) can outperform Graphics Processing Units (GPUs) due to their ability to execute highly parallel and customizable computing tasks efficiently. Recognizing this potential, this submission proposes leveraging Segment Routing traffic engineering (SR-TE) to intelligently direct specific artificial intelligence (AI) workloads to either FPGAs or GPUs based on traffic type, thereby optimizing performance and resource utilization.

Solar Panels Utility : Solar Compasses, Djamila Hassan Djibril Professor

Defensive Publications Series

With the development of new energy resources, solar panels are intended to be more energy-efficient, simple to install, and adaptable to changing environmental conditions. However, not all of the aforementioned statements are accurate they comprise a variety of layers that are combined and connected to a junction box that is connected to digital batteries that measure voltages charged by light consumption

in response to environmental changes or shifts in the sun's position brought on by climate change we decided to design a new utility Solar Compass , The goal of the new utility's design is position placement control, where it …

A Process For The Preparation Of Pacritinib Or Its Salts, Msn Laboratories Private Limited, R&D Center; Srinivasan Thirumalai Rajan, Revu Satyanarayana, Nomula Sathaiah And Rayavarapu Srinuvasa Rao

Defensive Publications Series

The present disclosure relates to a process for the preparation of Pacritinib and its intermediates. Pacritinib is represented by the following structural formula-1.

Formula-1.

Agentic Ai Framework For Semantic Transaction Data Compression And Retrieval, Palakh Shangle

Defensive Publications Series

The present disclosure relates to the field of Artificial Intelligence (AI), in particular to agentic AI framework for semantic transaction data compression and retrieval. The disclosed system comprises a plurality of AI agents, including an ingest agent, a selector agent, an encoder agent, a retriever agent and a governance agent, collectively configured to process large volumes of transaction records generated by electronic payment networks. In operation, transaction records are ingested, normalized, and analyzed to dynamically determine an appropriate compression policy based on storage efficiency, semantic fidelity and retrieval latency. Compressed transaction data is stored in an object storage system along …

Achieving Dynamic Urpf With Route Scale Optimizations, Anonymous

Defensive Publications Series

Systems and methods are disclosed for achieving dynamic Unicast Reverse Path Forwarding (uRPF) with route scale optimizations in an IP routing device, in which ingress traffic is sampled and analyzed to detect conditions indicative of source address spoofing or related attacks, and upon detecting such a condition, a control plane automatically enables uRPF at a selected scope including globally, per virtual routing and forwarding instance (VRF), and/or per interface, and subsequently disables uRPF once the attack condition subsides for at least a configured inactivity interval. Because enabling uRPF can reduce the maximum number of routes installable in a hardware forwarding …

Efficient Machine Learning And Serving Using Disjoint Hamiltonian Cycles, N/A

Defensive Publications Series

The technology described in this paper relates to the use of disjoint Hamiltonian cycles to support asymmetric communication patterns effectively while maintaining desirable latency control properties of direct-connect topologies. The technique used in this paper dynamically re-maps fixed disjoint dimensions of a direct-connect topology into a flexible set of disjoint Hamiltonian cycles that can trade network dimensions and the bandwidth-per-dimension to support asymmetric communication patterns. The mapping of the asymmetric network onto Hamiltonian cycles enables dividing the workload across Hamiltonian cycles of equal length.

Pipelined Execution Of Attention And Feed-Forward Network Operations On Heterogeneous Cores For Llm Decode Acceleration, N/A

Defensive Publications Series

The technology described in this paper relates to maximization of resource utilization for transformer based large language models (LLMs), specifically during an autoregressive decode phase. For the decode phase, each layer of a transformer model is primarily composed of two distinct computational blocks, an attention operator block and a feed-forward network (FFN) operator block. The attention operator block is overwhelmingly memory-bandwidth bound. In contrast, the FFN operator block is largely compute-bound, and the computations are dominated by large dense matrix multiplications. A chip architecture may be provided that contains at least two types of specialized co-located cores, a first core …

Large Language Model (Llm) – Agent System For Register Transfer Level (Rtl) Transformation, N/A

Defensive Publications Series

The technology described in this paper relates to transformation of functional register-transfer level (RTL) hardware designs for implementation by electronic design automation (EDA) physical synthesis tools. An intelligent, autonomous, and self-correcting RTL transformation system that is based on large language models (LLMs) is proposed. The system includes an LLM agent that is based on one or more LLM models, and a robust closed-loop transformation validation agent that is integrated with the LLM agent. A continuous feedback loop is used to allow the agent for RTL transformation validation to provide feedback to the LLM agent as error messages. The LLM interprets …

Coating Of Metallic Wires To Prevent Corrosion And Enhance Chemical And Mechanical Performance

Defensive Publications Series

This publication discloses methods and structures for applying corrosion‑resistant coatings to metallic wires and strips used as tensile and pressure armor elements in flexible pipe systems such as those defined in API 17J and API 15S (e.g., subsea, downhole, and annulus‑exposed environments). The coatings include fusion‑bonded epoxy (FBE), polyamide (e.g., PA11, PA12, and related thermoplastic polyamide families), and film/laminate constructs (e.g., XLPE with metallized or aluminum film layer, EVOH or PVDF-based multi layers), dual-layer systems (FBE primer + PA11 topcoat), nano-particle reinforced polymers (nanoclay-reinforced by PA11/PA12), and metallic-polymer hybrids (Zn-Al thermal spray combined with polymer overcoats). The disclosed approach focuses …

The Composite Virtual Content Viewing System That Provides An Immersive Virtual Content Observing Space For Greater Virtual Immersion., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a composite virtual content viewing system that provides an immersive virtual content observing space for greater virtual immersion. The composite virtual content viewing system that provides an immersive virtual content observing space for greater virtual immersion, contain a main structural part. The main structural part will contain a projection system for the left eye and a projection system for the right eye. The projection system for the left eye contains a collection of configurations that have the capability to effectively function as a collection of converging lenses, a collection of transparent configurations that have the capability to …

Packet Replication For Ai Training In Existing Data Centers, Zafar Ali, Pablo Camarillo, Clarence Filsfils, Ahmed Abdelsalam

Defensive Publications Series

Existing congestion management techniques for artificial intelligence (AI) data centers typically involve dedicated hardware for managing network congestion, which can be costly to implement, thereby requiring a large investment by network operators. Proposed herein is a technique to facilitate packet replication, which can make communications in an AI fabric resilient against congestion.

Avoiding Fabric Congestion Using An Sr-Te Agent For Ai Training Using Existing Data Centers, Zafar Ali, Clarence Filsfils, Francois Clad, Pablo Camarillo, Bruce Mcdougall

Defensive Publications Series

Existing approaches for managing congestion in an artificial intelligence (AI) data center (DC) network fabric typically involve dedicated hardware and are often reactive in nature. Proposed herein are techniques that can be utilized to avoid fabric congestion instead of building reactive techniques with new hardware. Specifically, a Segment Routing-Traffic Engineering (SR-TE) agent is proposed herein that can be utilized to avoid fabric congestion, which can enable enterprises to use their existing data centers for AI training.