arXiv subjects

Surender Suresh Kumar

Publications and source records attributed to Surender Suresh Kumar.

4 recordsLinked to original sources

arXiv2026

Evolving and Detecting Multi-Turn Deception using Geometric Signatures

Safety defenses for large language models (LLMs) are typically trained and evaluated on single-turn prompts, yet real attacks often unfold as indirect, multi-turn probing. To defend against this more nuanced form of deception, we present a unified pipeline that generates realistic multi-turn deceptive question sets via multi-objective genetic prompt optimization with co-evolving mutation operators. We validate this dataset through a human study, which also revealed that early generations yielded the most convincing deception and practical constraints such as adherence filtering and ordering effects. Using this data, we were able to detect deceptive attempts to access prohibited information using simple, explainable geometric signals in embedding space coupled with a lightweight feed-forward classifier. Three geometric features (angular coverage, distance ratio, and linearity) augmented with pairwise similarity statistics led to a compact predictive model that achieved consistently high recall (0.89) across base, reworded, and truncated (three-turn) scenarios, with test-time F1 ranging from 0.74-0.86. The results support a central hypothesis that multi-turn deceptive intent leaves a stable geometric footprint that enables lightweight, transparent screening without expensive end-to-end training. We further discuss responsible uses, limitations, and paths toward larger, more diverse human-evaluated datasets. The primary contribution to artificial intelligence is the multi-objective evolutionary framework for prompt generation, and the engineering application is the deployment of a lightweight geometric detection system for LLM safety infrastructure.

stat.ML↗

arXiv2026

Machine Learnability as a Measure of Order in Aperiodic Sequences

Research on the distribution of prime numbers has revealed a dual character: deterministic in definition yet exhibiting statistical behavior reminiscent of random processes. In this paper we show that it is possible to use an image-focused machine learning model to measure the comparative regularity of prime number fields at specific regions of an Ulam spiral. Specifically, we demonstrate that in pure accuracy terms, models trained on blocks extracted from regions of the spiral in the vicinity of 500m outperform models trained on blocks extracted from the region representing integers lower than 25m. This implies existence of more easily learnable order in the former region than in the latter. Moreover, a detailed breakdown of precision and recall scores seem to imply that the model is favouring a different approach to classification in different regions of the spiral, focusing more on identifying prime patterns at lower numbers and more on eliminating composites at higher numbers. This aligns with number theory conjectures suggesting that at higher orders of magnitude we should see diminishing noise in prime number distributions, with averages (density, AP equidistribution) coming to dominate, while local randomness regularises after scaling by log x. Taken together, these findings point toward an interesting possibility: that machine learning can serve as a new experimental instrument for number theory. Notably, the method shows potential 1 for investigating the patterns in strong and weak primes for cryptographic purposes.

cs.LG↗

arXiv2026

MirrorMark: Generalizable Mirrored Sampling for Multi-bit LLM Watermarking

As large language models (LLMs) become integral to applications such as question answering and content creation, reliable content attribution has become increasingly important. Watermarking is a promising approach, but most existing methods either provide only binary signals or achieve multi-bit embedding by distorting the generation distribution. We propose MirrorMark, a generalizable mapping-centric approach for multi-bit LLM watermarking. MirrorMark separates the symbol mapping rule from the base watermarking sampler and maps each symbol to a mod-1 mirroring transformation of a detector-reproducible pseudorandom object, such as sampling values or permutation ranks. A binary-tokenizer analysis shows that complementary mappings yield larger matched--mismatched score gaps than independent-key or shift-based mappings. When composed with a distortion-free base sampler, MirrorMark preserves the token probability distribution by design and maintains text quality in practice. To support practical payload embedding, we introduce a Context-Anchored Balanced Scheduler (CABS), which balances token assignments across message positions while localizing edit effects. We further provide theoretical EER analyses for two representative sampler instantiations. Experiments show that MirrorMark achieves strong detectability and bit accuracy while maintaining text quality comparable to non-watermarked generation.

cs.CR↗

arXiv2026

Survival is the Only Reward: Sustainable Self-Training Through Environment-Mediated Selection

Self-training systems often degenerate due to the lack of an external criterion for judging data quality, leading to reward hacking and semantic drift. This paper provides a proof-of-concept system architecture for stable self-training under sparse external feedback and bounded memory, and empirically characterises its learning dynamics and failure modes. We introduce a self-training architecture in which learning is mediated exclusively by environmental viability, rather than by reward, objective functions, or externally defined fitness criteria. Candidate behaviours are executed under real resource constraints, and only those whose environmental effects both persist and preserve the possibility of future interaction are propagated. The environment does not provide semantic feedback, dense rewards, or task-specific supervision; selection operates solely through differential survival of behaviours as world-altering events, making proxy optimisation impossible and rendering reward-hacking evolutionarily unstable. Analysis of semantic dynamics shows that improvement arises primarily through the persistence of effective and repeatable strategies under a regime of consolidation and pruning, a paradigm we refer to as negative-space learning (NSL), and that models develop meta-learning strategies (such as deliberate experimental failure in order to elicit informative error messages) without explicit instruction. This work establishes that environment-grounded selection enables sustainable open-ended self-improvement, offering a viable path toward more robust and generalisable autonomous systems without reliance on human-curated data or complex reward shaping.

cs.AI↗