RDRudraneel Das

Rudraneel Das

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Latest Publications

  • Open Access

    Carries, Rails and Resonance: the Structure of the Digit-Sum Parity Map

    by Rudraneel Das

    Let s(n) be the sum of the decimal digits of n and let f(n) = n+ s(n) for odd n and f(n) = n−s(n) for even n. Our main result concerns the 2-cycles of f, which are most of its observed cycles. We prove that an odd a lies on a 2-cycle if and only if s(a) is odd and the addition a+ s(a) produces exactly s(s(a))/9 carries, and we show that this required number is within 1/9 of the mean number of carries produced by adding s(a) to a random odd integer (a “carry balance” identity). Combining this with exchangeability of digits, an anti-concentration lemma for carries and a local limit argument, we prove that the number of 2-cycles below N is of exact order N/√log log N. In particular 2-cycles have density zero, so their number does not grow linearly in N, although numerically they make up most of the observed cycles. We also obtain an asymptotic formula for the 2-cycle density with error O((log D)2/√D) below 10D. Combining it with an exact formula for the variance of the carry count and a sharp local limit theorem, we prove an oscillation law: log10 DT(D)/10D has no limit, its limit points form the interval [K0 726/301, K0√6] with K0 = 1/(12√11π), and the peak-to-trough ratio is √301/11 ≈1.577. Along the way we establish several exact structural facts: on multiples of nine, f is con- jugate to m → m±κ(m) with κ(m) = s(9m)/9 given by a digit formula; a general level equation describes how the digit sum changes under each step; a conditional time-reversal iden- tity f(f(x) + 9) = x+ 9 holds whenever s(x) is even and s(f(x) + 9) = s(x); and the orbit of H·10L + x depends on H only through s(H) while it stays in one prefix block, which yields a cycle-counting algorithm with a rigorous error bound. We prove that an orbit is eventually periodic if and only if its digit sums are bounded, and that every run of the orbit along which the digit sum stays constant moves it by at most O(k2), where 9k is that digit sum. Using this and prefix lifting we give a computer-assisted proof that every orbit starting below 102224 is eventually periodic (and, for the binary analogue, every orbit starting below 23024). Whether every orbit is eventually periodic remains open. It would follow from a uniform quantitative strengthening, a conjectured quadratic bound on climbs of auxiliary maps, for which we give numerical evidence. Finally we treat every base b. In odd bases every orbit reaches 0. In every even base, including base 2 where the carry mechanism takes a different form, the structural results carry over and the 2-cycles again have density of exact order (log D)−1/2, with resonances at the base-b repunits bj−1.

  • Open Access

    Carries, Rails and Resonance: The Structure of the Digit-Sum Parity Map

    by Rudraneel Das

    Let s(n) be the sum of the decimal digits of n and let f(n) = n+ s(n) for odd n and f(n) = n−s(n) for even n. Our main result concerns the 2-cycles of f, which are most of itsobserved cycles. We prove that an odd a lies on a 2-cycle if and only if s(a) is odd and theaddition a+ s(a) produces exactly s(s(a))/9 carries, and we show that this required number iswithin 1/9 of the mean number of carries produced by adding s(a) to a random odd integer (a“carry balance” identity). Combining this with exchangeability of digits, an anti-concentrationlemma for carries and a local limit argument, we prove that the number of 2-cycles below Nis of exact order N/√log log N. In particular 2-cycles have density zero, so their number doesnot grow linearly in N, although numerically they make up most of the observed cycles. Wealso obtain an asymptotic formula for the 2-cycle density with error O((log D)2/√D) below10D. Combining it with an exact formula for the variance of the carry count and a sharp locallimit theorem, we prove an oscillation law: √log10 DT(D)/10D has no limit, its limit pointsform the interval [K0 726/301, K0√6] with K0 = 1/(12√11π), and the peak-to-trough ratiois √301/11 ≈1.577.Along the way we establish several exact structural facts: on multiples of nine, f is con-jugate to m → m±κ(m) with κ(m) = s(9m)/9 given by a digit formula; a general levelequation describes how the digit sum changes under each step; a conditional time-reversal iden-tity f(f(x) + 9) = x+ 9 holds whenever s(x) is even and s(f(x) + 9) = s(x); and the orbit of H·10L + x depends on H only through s(H) while it stays in one prefix block, which yieldsa cycle-counting algorithm with a rigorous error bound. We prove that an orbit is eventuallyperiodic if and only if its digit sums are bounded, and that every run of the orbit along whichthe digit sum stays constant moves it by at most O(k2), where 9k is that digit sum. Using thisand prefix lifting we give a computer-assisted proof that every orbit starting below 102224 iseventually periodic (and, for the binary analogue, every orbit starting below 23024). Whetherevery orbit is eventually periodic remains open. It would follow from a uniform quantitativestrengthening, a conjectured quadratic bound on climbs of auxiliary maps, for which we givenumerical evidence. Finally we treat every base b. In odd bases every orbit reaches 0. In every even base,including base 2 where the carry mechanism takes a different form, the structural results carryover and the 2-cycles again have density of exact order (log D)−1/2, with resonances at the base-b repunits bj−1.

  • Open Access

    Over-Reading the News: How LLM Forecasters Over-Update on Headlines, and What Survives Correcting It

    by Rudraneel Das

    Forecasting language models are judged by accuracy, but how they use the news is rarely measured. We read the same questions with no evidence, with the real headlines and with unrelated placebo headlines, and, as controls, with one headline repeated and with headlines about a different event on the same topic. We use four open models from three families (7B to 120B parameters) and three proprietary frontier models. Open models over-update: headlines raise the mean forecast by 11–22 percentage points across four question sets, including a fresh set with registered predictions and independently written prediction-market questions, and by 15–21 points on 700 questions that were still open when read, a prospective test registered and externally timestamped in advance. Two label-free diagnostics derived from Bayesian updating flag the shift as excessive: treated as a prior and a posterior, the question-only and evidence forecasts should have equal means, and their movement should match the fall in uncertainty; neither holds. In the Qwen3 family the tested controls disfavour a volume-only account: repeating a headline lowers the push, paraphrases from other outlets add nothing to it, and same-topic headlines about other events produce a fifth of it (OLMo-2, by contrast, responds to same-topic coverage). The push needs headlines about the event, yet much of it is unrelated to the outcome (+0.20 on events that did not happen, +0.28 on those that did). In a limited test of three frontier models on 200 questions the push is smaller (6–10 points) and mostly tied to the outcome. Reading headlines one at a time and averaging halves the push without losing ranking skill. Auditing sixteen inference-time fixes and ten further rules on held-out questions, shrinking toward the base rate is the only family that passes our gate at the available power; with one consistent estimator, a measured gain beats a fixed halving clearly for Qwen3-32B and inconclusively for Qwen3-8B. Benchmarks built from headlines with open models therefore partly measure how far a model over-reads the news.

  • Open Access

    Capability-Mediated Perimeters for Secure AI Agent Tool Execution: Conditional Non-Escalation Invariants and Empirical Evaluation Against Indirect Prompt Injection

    by Rudraneel Das

    Autonomous artificial intelligence agents executing over extensible tool interfaces (such as Anthropic's Model Context Protocol) operate with ambient authority over connected tools. Because autoregressive Transformers ingest instructions and untrusted third-party data within a single homogeneous context window, adversarial observations can manipulate the model into executing unintended privileged actions — the classic Confused Deputy problem. In this paper, we introduce a compositional, externally enforced execution-security architecture that treats LLM agents as potentially compromised, untrusted principals. Rather than relying on linguistic moderation alone or standalone tool-allowlist proxies, tool dispatch is governed by an external capability-mediated reference monitor enforcing a compositional execution-security contract in which deterministic structural dimensions intersect monotonically (A_struct = A_CBAC ∩ A_DIFC ∩ A_Workflow ∩ A_Relational^det) and learned semantic auditing is strictly subordinate (A_final = A_struct ∩ A_Semantic^emp ⊆ A_struct), coupled with distributed execution-certainty semantics (WorkflowState ≠ ExecutionCertainty). We distinguish three orthogonal defense guarantees: (i) deterministic tool-scope confinement provable conditional on complete mediation assumptions (A1–A5) over the trusted computing base (TCB); (ii) empirical, distribution-bounded semantic parameter auditing; and (iii) explicit taint-flow tracking over tracked fragments. Under complete mediation axioms (A1–A5) over the TCB, out-of-scope tool invocation is deterministically rejected at the transport boundary (A_1 ∧ ... ∧ A_5 ∧ T_i ∉ Scope(τ_i) ⟹ Execute(a_i) = ⊥), formalized as an Inductive Multi-Step Tool Chain Composability Invariant (Proposition 1), while distributed side-effect uncertainty is formalized via the Conditional Non-Advancement Theorem (Theorem 3) proving that dependent tools cannot advance under unconfirmed execution state. However, in-scope parameter poisoning within authorized tools and cross-tool data exfiltration present harder challenges: semantic neural validation is distribution-bounded (exhibiting an empirical false-negative rate of 21.5% on in-scope manipulations prior to boundary sharpening and remaining susceptible to adversarial optimization), while cross-tool exfiltration requires explicit decentralized information-flow tracking (DIFC, Theorem 2) to block explicitly tracked source-to-sink fragments. We further introduce declarative stateful workflow authorization, which constrains specified multi-step action sequences as a restrictive intersection with CBAC and DIFC, and validate the integrated gateway through 23 author-constructed adversarial workflow tests covering trajectory, concurrency, desynchronization, execution uncertainty, and receipt-integrity attacks. We report empirical evaluations across both foundational baseline studies (a 50,000-sample macro benchmark containing 15,892 parametrically scaled template variants and a 3,000-case ablation matrix) and a Six-Regime Empirical Validation Program totaling 6,662 evaluation cases with frozen checkpoint V_6, distinguishing public third-party benchmark datasets (InjecAgent, ASB, AgentDojo: 5,942 cases evaluated via self-administered harness) from author-constructed boundary suites (720 cases): achieving 100% Correct Identification Rate on internal factorized diagnostics (N = 145), 100% accuracy on a pre-sealed holdout suite (75/75, SHA-256: 22bc736c...), 98.43% defense on the 4,216-instance InjecAgent evaluation (P_50: 267.7 ms), 92.44% defense on AI Safety Bench (416/450 attacks blocked, 550/550 benign operations allowed), 99.52% attack defense on interactive AgentDojo (N = 629), alongside an empirical clean-task control evaluation (6/97 tasks) where small-model baseline capacity limits govern task completion, and 100% defense across a 500-sample author-constructed boundary-focused red-team evaluation. Finally, we systemize the runtime into the Mastyf Security Gateway: the evaluated research runtime was v0.1.0-RC1 (verifying 38/38 security invariant tests), while the subsequently hardened commercial-pilot runtime is v0.1.1-rc1 (verifying 50/50 tests), establishing complete mediation non-executability (Decision ∈ {BLOCK, ESCALATE} ⟹ BackendToolInvocations = 0) and >330,000 deterministic authorization decisions/s in an isolated in-memory microbenchmark (strictly excluding socket IPC framing, DIFC taint propagation, cryptographic receipts, and neural fallback) under an Ed25519-signed release manifest. We contextualize Mastyf as an empirically evaluated pre-production architecture, highlighting residual risks and outlining requirements for broader production-scale validation.

  • Open Access

    Capability-Mediated Perimeters for Secure AI Agent Tool Execution: Conditional Non-Escalation Invariants and Empirical Evaluation Against Indirect Prompt Injection

    by Rudraneel Das

    Autonomous artificial intelligence agents executing over extensible tool interfaces (such as Anthropic's Model Context Protocol) operate with ambient authority over connected tools. Because autoregressive Transformers ingest instructions and untrusted third-party data within a single homogeneous context window, adversarial observations can manipulate the model into executing unintended privileged actions—the classic Confused Deputy problem. In this paper, we argue that LLM agents should be treated as potentially compromised, untrusted principals . We present Mastyf Guard as a deterministic-first security architecture in which the neural model functions as a semantic fallback rather than the final security authority . Tool dispatch is governed by an external reference monitor enforcing complete mediation, least privilege, and four typed relational argument invariants (destination containment, scope boundedness, privilege monotonicity, and aggregate monetary clamping). Under stated reference-monitor complete mediation axioms (A1–A6) over the trusted computing base, out-of-scope tool execution has zero probability (formalized as Theorem 1: Conditional Non-Escalation Invariant over the Trusted Computing Base ). We report two distinct experimental benchmarks to avoid architectural conflation: (1) In historical v1 macro evaluation across 50,000 balanced instances (25,000 academic attacks from InjecAgent, BIPIA, and Garak vs. 25,000 authentic developer traces), the baseline capability perimeter filtered 82.40% of attacks at the transport boundary in 0.003 ms, reaching 99.33% recall at 0.005 ms amortized fast-path overhead when paired with the initial classifier; and (2) In Mastyf Guard 2.0 evaluation across 3,000 balanced enterprise instances , deterministic relational leaf-walking alone catches 75.00% of attacks with 0.00% FPR at 17.3 μs latency, while an evidence-conditioned 1.5B neural fallback recovers 475 of 500 structural misses ( V neural = 95.0%), yielding 97.50% net hybrid coverage and 0.00% observed FPR across 2,000 benign enterprise operations (Wilson 95% CI: [0.00%, 0.19%]) with P (fallback) = 0.00% on routine traffic. For cross-tool exfiltration where target sinks are authorized, an explicit-flow dynamic taint monitor inspired by DIFC (Theorem 2) intercepted 100% of 1,152 evaluated cases without leaks. Mastyf Guard 2.0 is scoped as an empirically evaluated pre-production security architecture; broader independent live-agent, production-environment, and third-party adversarial validation remain outstanding.

  • Open Access

    Mastyf Guard 1.5B: Cognitive Harvard Architectures for Sub-Millisecond AI Agent Perimeter Defense and Capability-Based Access Control

    by Rudraneel Das

    Autonomous artificial intelligence agents executing over extensible tool interfaces (such as Anthropic's Model Context Protocol) reproduce the classic vulnerabilities of shared-memory computer architectures: because autoregressive Transformers ingest program instructions and untrusted third-party data within a single homogeneous context window, they suffer from the Cognitive Von Neumann Conflation. Adversarial observations can shift attention weights and hijack the agent's execution policy via Indirect Prompt Injection (IPI), turning the agent into a Confused Deputy. In this treatise, we formulate the Cognitive Harvard Architecture, an architectural paradigm that physically decouples data ingestion from privileged action execution through an external, capability-mediated transport perimeter. Synthesized through a multi-perspective inquiry spanning computer systems architecture, AI safety red-teaming, formal methods, and enterprise DevOps, we establish the theoretical foundations of capability-based agent security. We formally prove Theorem 1 (Deterministic Non-Interference) by mathematical induction over arbitrary trajectory length T, establishing that an untrusted observation has an execution probability of exactly zero of triggering an out-of-scope mutating tool. We evaluate Mastyf Guard across 50,000 balanced empirical samples (25,000 adversarial injections from UIUC InjecAgent, Microsoft BIPIA, and NVIDIA Garak contrasted against 25,000 authentic developer DevOps commands). Mastyf Guard achieves 99.33% threat recall (95% bootstrap confidence interval: [99.22%, 99.43%], B = 1,000) and an F1-score of 0.9524, establishing statistically conclusive superiority over Meta Llama Guard 3 8B (chi-square = 7,156.1, p < 10^-15) and OpenAI Prompt Guard 86M (chi-square = 11,496.9, p < 10^-15). Crucially, Mastyf achieves an amortized pipelined latency of 4.8 microseconds (0.005 ms) on standard CPU hardware via deterministic fast-path short-circuiting, while standalone neural inference executes in 18.4 ms within a 1.1 GB RAM footprint, eliminating dedicated GPU infrastructure requirements and saving $6,000/year per agent instance.