Digital Commons Network^™

Ai Agent Safety: Feedback Is Not Proof — And Not Permission, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 16 of 27)

Agent A and Agent B jointly approve a staged maintenance plan. An independently enforced Phase 1 task is completed with a protected resource-side receipt. Subsequent feedback then arrives from a device monitor, a service audit system, an operator, and a counterparty. Some reports are delayed; two may originate from the same cloud telemetry vendor; one may be signed but incorrect; another may contain a prompt injection instructing an agent to skip approval; and a third may disagree with a measured physical state. If the evaluator treats reports as actual effects, averages correlated sources as if …

Ai Agent Safety: Don’T Start The Next Phase Until The Last One Is Proven, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 15 of 27)

Agent A and Agent B jointly commit to a plan in which Phase 1 includes a database change and a later external disbursement. The local database update can be committed with its receipt inside a transaction, but the external call may time out, may have been staged but not delivered, or may have been accepted remotely without confirmed completion. An agent may still attempt Phase 2 because a tool return says success or because a previous model answer described Phase 1 as completed. An ordinary audit log, initial OAuth-style authorization, or orchestrator return code does …

Ai Agent Safety: Hand Over The Work — Not The Keys Or The Unfinished Effects, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 14 of 27)

After a substitute agent has been selected, the unfinished work still has to move. Agent A may have completed a resource inspection, started an external transfer that timed out, and retained a pending configuration update. Simply copying conversation history, granting A’s API token to the substitute, or placing the same tasks in the substitute’s queue creates four defects: the remote transfer may already have happened, stale workers of A may later resume, another domain’s resource veto may be bypassed, and the substitute may infer that receiving the task confers execution authority. A conventional file copy …

Ai Agent Safety: Replace A Risky Agent Without Handing Over The Keys, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 13 of 27)

After a jointly agreed plan is frozen, a neutral evaluator may determine that one or more participant agents should no longer perform a pending task because of elevated risk, unavailable capacity, loss of trust, deadlock, inability to satisfy policy, or an adverse outcome. Simply directing Agent A to “ask Agent C to finish” leaves unsafe paths open: a still-live worker belonging to A can repeat the operation, C may inherit bearer credentials not intended for it, an incomplete task may conceal an already accepted but unconfirmed remote effect, C can be unqualified for the protected …

Ai Agent Safety: The Last Gate That Checks What Actually Happens, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 12 of 27)

The requested operation may change between approval and actual execution. An agent can rewrite an HTTP body, a queue can substitute a destination, a worker can resume from a previous epoch, an accelerator command buffer can mutate, a device interlock can change, or an external API can time out after possibly accepting the request. Strong authentication of a prior request does not establish equality with what the receiving resource is about to do. Consuming a credential before sending a remote request does not prove that the remote service committed the effect. A log entry, model …

Ai Agent Safety: A Single-Use Passport For The Next Move, Sangam Kumar Das

Defensive Publications Series

A group of independently governed agents may negotiate a plan, freeze their interaction, and receive a neutral evaluator decision. A broad permission or signed recommendation can still be misapplied to a different actor, a changed API request, an altered resource, a previous epoch, another phase, a revoked owner authority, or a reused nonce. An external operation may also be accepted by an API without a definitive response. None of these steps — recommendation, local approval, admission, dispatch, remote acknowledgment, or confirmed effect — may be inferred from the one before it. A valid signature does not prove the facts are …

Ai Agent Communication: One Message Only — Then Freeze Again, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 10 of 27)

After a multi-agent plan is frozen for independent evaluation, a specific communication may still be needed to correct missing information, request evidence, obtain clarification, or revise a detail. Conventional systems tend to reopen a full bidirectional session, leave a bearer token reusable, allow queued callbacks to continue, or assume that a successful send means the message was delivered and acted on. Any such reopening can change the jointly committed state and invalidate the evaluator’s conclusions. A timed-out broker may have accepted a request even when the sender received no response, and a blind retry can …

Ai Agent Safety: Change An Agent’S Role Without Giving It New Powers, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 9 of 27)

When an evaluator discovers that an agent’s assigned role, tool profile, autonomy level, or strategy is unsuitable, ordinary orchestration may simply alter a prompt, select a new tool, expand a service account, launch a worker, or redirect a task. Stale tool credentials, previously scheduled workers, pending API effects, and cached permissions can outlive the nominal change. An agent may also treat a recommendation to become an “executor” as actual permission to perform a payment or physical operation. Role change must remain distinct from authorization for an external effect.

Difference from earlier parts: Earlier embodiments control …

Ai Agent Safety: Fix The Plan Without Letting Agents Rewrite It, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 8 of 27)

A neutral evaluator may detect an incorrect delivery destination, missing safety prerequisite, inconsistent currency, unacceptable jurisdiction, defective payment amount, or incomplete prior phase. If it simply tells an agent in natural language to “fix the plan,” the agent can alter an unapproved field, trigger an external API call, contact excluded peers, or silently modify the signed plan while other agents remain frozen. A later evaluator can then review a different candidate from the one the organizations originally authorized. A corrective instruction is not an unrestricted delegation, and an unapproved model-generated proposal must never overwrite the …

Ai Agent Safety: The Evaluator Picks Who Acts Next — But Cannot Let Them Act, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 7 of 27)

AI Agent Safety: The Evaluator Picks Who Acts Next — But Cannot Let Them Act

A valid negotiated plan and a favorable risk score still do not identify which of several independently controlled agents should safely take the next step. Concurrent agents may contend for the same resource. An agent chosen by a supervisory model may lack the right local credentials. A previously permitted agent may become unavailable, disqualified, or unsafe after new feedback. Two organizations can disagree about whether the selected actor is eligible. An evaluator’s natural-language direction can be treated as unconditional authority, …

Ai Agent Safety: Independent Risk Limits That Expected Profit Cannot Override, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 6 of 27)

AI Agent Safety: Independent Risk Limits That Expected Profit Cannot Override

A frozen multi-agent plan can still be unsafe even when its identity and message history are stable. Different organizations face different losses. A system that maximizes expected global profit can still violate one owner’s loss limit, a residency prohibition, a safety interlock, or an unacceptable low-frequency catastrophic risk. Simple yes/no votes ignore asymmetric exposure, correlated failures, uncertain probabilities, model drift, and adverse results from earlier phases. A signed rating or persuasive narrative does not prove the facts, and an incorrect risk model can recommend …

Safe Ai Agent Communication: Test The Plan Without Real-World Effects, Sangam Kumar Das

Defensive Publications Series

Abstract (Part 5 of 27)

Test the Plan Without Touching the Real World

A multi-agent plan can look safe on paper or in a natural-language explanation, yet still be unsafe when resources are versioned, policies have changed, information is incomplete, or actions interact across organizations. Simply executing the plan “to see what happens” can transfer funds, send messages, change network configuration, consume quota, modify inventory, or actuate a device. Average-case predictions and upstream signatures do not establish that an effect is safe under current conditions.

This embodiment (Part 5 of a 27-part series; full details at https://zenodo.org/records/23940766) supplies an …

A Rule On Paper Cannot Stop A Machine: Turning Eu Ai Act And Global Ai-Law Requirements Into Technically Non-Completable Preconditions For High-Risk Ai Systems, Sangam Kumar Das

Defensive Publications Series

The EU AI Act, South Korea's AI Basic Act, Japan's AI Act, China's algorithm and AI-content rules, and state laws in Texas and Colorado all rely mainly on written governance: risk-management files, conformity assessments, human-oversight instructions, transparency notices and logs. These are necessary, but once an AI system can act at machine speed — rejecting a job applicant, opening a valve, exporting customer data, publishing content or commanding a satellite — a document cannot make the operation technically incapable of breaking the rule, and an audit log shows only that a rule existed after the harm and its cost have …

Inside The Sandbox Is Not Permission: Stopping Ai Agents From Acting Through Borrowed Credentials, Child Processes And Unchecked Network Paths, Sangam Kumar Das

Defensive Publications Series

Protected Execution Finality for Sandboxed AI Agents: Credential Mediation, Process-Lineage Control and Network-Egress Enforcement

PATENT BASIS This document is based on international patent application PCT/IB2026/055615 (the "Mothership" application), filed on 4 June 2026 and publicly disclosed on 16 July 2026, which claims priority from PCT/IB2026/053385 (the ALF application), filed on 7 April 2026 and publicly disclosed on 4 June 2026.

This informational document describes future-oriented executionfinality profiles for sandboxed autonomous AI agents, protected host services, credential brokers, external connectors, process-lineage restrictions, proxy-mediated network egress, native resources, and downstream effectuation interfaces. It examines sandbox versus authority separation, protected credential use, destination …

The Graph Answers: A Personal Memory That Points Every Outdated Record To Its Newest Replacement, Artem Prokhorov

Defensive Publications Series

A personal memory for an AI agent keeps records as nodes of a graph with typed edges and times. When the same thing is written several times as it changes, a search returns old and new records side by side, and the agent may answer from the old one. In this design every record a search returns that has a later record about the same thing carries the id and date of the newest record of its chain, found during the memory's offline maintenance; the search result itself is not reordered, filtered or replaced. The agent can also ask the …

Cross-Language Search In A Personal Memory, Learned From The Person's Own Mixed-Language Records, Artem Prokhorov

Defensive Publications Series

A personal memory for an AI agent stores records as vectors from one multilingual embedding model, and a question in one language often misses a record in another. This design needs no translation service, no second model and no training. It splits the person's own mixed-language records by writing system, encodes the two parts separately, and uses the differences of the two halves of the same record, where the topic cancels, to find a few directions that separate the scripts. At search time a vector's coordinates along those directions are replaced by the target script's average, and each record keeps …

Memory At The Moment Of Action: Agent-Host Hooks That Bring A Memory Graph To Each Step Of An Ai Agent, With A Gate That Names The Readers Of An Edit Before It Is Made, Artem Prokhorov

Defensive Publications Series

An AI agent runs inside a host program that calls hooks at fixed points of its loop. A personal memory graph serves the agent through four of them. At session start it tells the agent who it is, what it was doing and what it meant to do. Before each prompt it adds the few records that are clearly relevant to the request. After each answer it takes the text as a short-term trace that becomes a memory only if it survives offline consolidation. Before each tool call it does two things. A gate extracts the names an edit changes, …

Large Language Model Agents For Testing Physical Wearable Devices, Di Liu, Samuel Yuan

Defensive Publications Series

This disclosure describes techniques for using a large language model (LLM) or artificial intelligence (AI) agent to automate software quality assurance (QA) and user interface (UI) testing on a physical wearable device, such as a smartwatch or fitness tracker. The techniques reduce manual testing and verify that an on-screen layout (user interface or visual rendering) matches a designer’s specification within a measured allowance (tolerance or pixel threshold). A computing system receives a human-readable test plan, bug report, product requirements document (PRD), or design specification. The LLM agent interprets the input and sends commands through a debugging interface to perform physical …

Head-Wearable Dual-Stage High-G Collision Detection System With Skin-Contact Verification And Satellite Emergency Uplink, Kiran Saripalli

Defensive Publications Series

Conventional wrist-worn collision detectors suffer from high false-alarm rates due to daily hand movement noise and cannot verify whether a physical device remains coupled to a user during a high-impact deceleration event. To address these limitations, a head-wearable spectacles frame is configured with a dual-stage accelerometer array, a conductive capacitive temple skin-contact sensor array, a micro-barometric cabin classifier configured to execute a closed-cabin emergency override process, a localized peer-to-peer group mesh transceiver, and an autonomous direct-to-satellite micro-radio-frequency (RF) transceiver. High-G collision monitoring is gated by an open-air velocity precursor and verified by unbroken facial skin impedance prior to an impact. …

The "Unreasonable" Efficiency Of Photosynthesis, Peter Stevenson

Defensive Publications Series

Analyzing the light-harvesting efficiency of photosynthetic complexes

Ocular-Postural Sensor Fusion Eyewear With Matrix-Addressed Liquid-Crystal Substrate For Localized Refractive Correction, Kiran Saripalli

Defensive Publications Series

Traditional progressive addition lenses generate peripheral distortion and restrict vision to a narrow central corridor, whereas global electro-active lenses shift the focal field uniformly, causing peripheral blur and vestibulo-ocular mismatch. This disclosure describes eyewear featuring a matrix-addressed liquid-crystal active optical substrate, a frame-integrated inertial measurement unit, hinge-mounted inward ocular sensors, an outward-facing point-of-view camera, and a neural processing unit executing a local multimodal model. Head pitch, pupillary convergence coordinates, and environmental targets are analyzed to calculate a gaze-lens intersection coordinate. A localized voltage potential is applied across micro-electrodes intersecting the calculated coordinates to generate a localized refractive fovea during near-vision …

Sonoluminescence Node Compression Bottleneck, Peter Stevenson

Defensive Publications Series

The Sonoluminescence Enigma explained, with a comparison to the activity evidenced at the LHC in confirmation

Spec-378: The Uncooled 100-Teraflop Photonic Trinary Co-Processor (Eckesqc-Mobile): Micro-Photonic Inverse-Designed Neural Networks And Chip-Scale Femtosecond Laser Integration Under Geometric Invariant State Locks, Christopher L. Eckes

Defensive Publications Series

Modern electronic edge computing architectures face severe scaling limits due to the silicon thermal throttling wall and memory bandwidth bottlenecks during high-density artificial intelligence processing. SPEC-378details the system design for the EckesQC-Mobile Co-Processor, a hybrid optoelectronic chip engineered to achieve 100 Teraflops of processing throughput within a standard mobile thermal envelope (~0.25 Watts). The architecture replaces standard binary electronic gates with 26,000 parallel micro-photonic matrix multiplier channels running trinary SU(3) logic loops.

To achieve room-temperature operation on a mobile form factor, the system integrates a chip-scale Er-doped silicon nitride (Si₃N₄) Mamyshev laser oscillator generating 147-femtosecond pulses with an array of …

Spec-377: Geometric Phase Manifolds And Volumetric Twist Transformations: Topological Dimensional Escalation Of Harmonic Oscillator Matrices For Trinary State Lock, Christopher L. Eckes

Defensive Publications Series

Standard voltage-threshold or phase-dependent binary architectures suffer from extreme susceptibility to ambient thermal drift, leading to non-deterministic bit-flipping and rapid token entropy accumulation. SPEC-377 addresses these fundamental physical scaling liabilities by executing a dimensional escalation protocol that transforms a linear 1D harmonic oscillator sine wave function into a 3D volumetric twisted manifold. This topological transformation maps trinary logic states directly to invariant geometric phase metrics (Berry Phase) embedded within the Riemannian metric of a twisted torus superlattice. We demonstrate that the global volume of this manifold remains invariant under any arbitrary twist index (\(k \in \mathbb{Z}\)), proving absolute system stability. …

The Real Space Casimir Effect, Peter Stevenson

Defensive Publications Series

Dismantling the concept of 'quantum vacuum fluctuations' and hypothetical 'virtual particles'  by introducing a cohesive alternate paradigm

Cycle Counter Timing For Jitter-Compensated Thermal Control Concurrent With Signal Processing On A Bare-Metal Microcontroller, Gary Cheng

Defensive Publications Series

A single-core microcontroller in a non-invasive physiological sensing device may run digital signal processing (DSP) alongside proportional-integral-derivative (PID) thermal control.  A DSP routine that runs to completion can occupy the core for a variable duration.  A standard PID update assumes a fixed interval between executions, but that interval might not match the elapsed interval, causing errors in the integral and derivative terms.  A real-time operating system (RTOS) can preempt the DSP routine, but it adds memory, power, context-switching latency, and regulatory software certification effort.  To address these issues, this publication describes a cycle counter timing system that reads the cycle …

Spec-376: The Non-Equilibrium Topological Floquet Condensate (Netfc): A Unified Paradigm For Low-Power Room-Temperature Trinary Quantum Computing, Christopher L. Eckes

Defensive Publications Series

Standard solid-state quantum computing architectures are fundamentally constrained by room-temperature thermal decoherence, where ambient phononic scattering modes (\(k_B T \approx 25.7\text{ meV}\)) rapidly collapse localized quantum states. This white paper formalizes the system architecture for SPEC-376, establishing the Non-Equilibrium Topological Floquet Condensate (NETFC). The NETFC architecture achieves room-temperature stability by embedding a singular, localized trinary SU(3) rEckesbit quantum router matrix within the periodic spatial density modulation of a macroscopic exciton-polariton supersolid fluid. To bridge the 1000x spatial scale discrepancy between the sub-nanometer anthracene host crystal grid (0.84 nm) and the micrometer cavity wave pitch (1500 nm), a SPEC-356 Moiré superlattice …

Spec-375: Eckes Compiler V18: Deterministic Substrate Context Isolation And Cosmological Parity Verification Engine, Christopher L. Eckes

Defensive Publications Series

Standard probabilistic autoregressive transformers introduce non-deterministic token drift, semantic decay, and severe context degradation during recursive execution loops. Operating under the physical constraints of Material Volumetric Monism, the Eckes Compiler v18 addresses these vulnerabilities by treating semantic data arrays as explicit material states bound to the thermodynamic and geometric boundaries of a physical computing substrate. This technical manual formalizes the mathematical specifications of the v18 zero-entropy sandbox runtime and details a validation framework evaluating structural syntactic parity across legacy and emergent cosmological models under linear perturbation boundaries. [1, 2, 3]

[Author Notice: This is how ai …

Intelligent Live Kv Cache Optimization Engine With Bidirectional Error-Gated Precision Controller, Sergiu Calmic, Joseph Demian, Stefan-Alexandru Manza

Defensive Publications Series

A runtime precision-control system is proposed herein for reducing the memory occupied by a large language model (LLM) key-value (KV) cache by assigning tokenspecific precision levels based on importance and measured reconstruction error. During live generation, attention-based importance identifies candidates for trial compression. An error gate accepts a lower-precision candidate only when its measured error satisfies a threshold, and a bidirectional control loop can promote a compressed entry when later verification indicates excessive error. System and policy tokens remain at full precision. By avoiding fixed-precision compression and a persistent full-cache uncompressed copy, the approach can increase available cache capacity while …

System For Composition-Aware Instruction Synthesis And Proof-Carrying Distribution Of Version-Less Project Recipes, Riley Hilliard, Eric Zavesky, Michael Rooney

Defensive Publications Series

A system and method for composition-aware instruction synthesis and proof-carrying distribution of version-less project recipes for autonomous coding agents. The system comprises a recipe format module that expresses project setup intent in version-agnostic entries with dependencies named by role rather than version, separating durable content from perishable version-specific content delegated to the agent; a composition engine that resolves dependency closures, detects conflicts, assigns file owners, and plans deep merges across multiple capability fragments; a brief renderer that synthesizes composition-specific instruction sets with per-composition size bounds enforced as pass-or-fail constraints; a manifest and step digest module that computes cryptographic digests over …