Kubernetes-Native Model Context Protocol RAG Server for Adversarial AI Research
Complete technical documentation of designing, implementing, and deploying a Kubernetes-native Retrieval-Augmented Generation server implementing the Model Context Protocol for adversarial AI research. Addresses indexing 5,395 offensive security documents with MITRE ATT&CK mappings, overcoming Linux container security restrictions blocking asyncio syscalls (socketpair() problem), migrating from FastAPI to Flask/Gunicorn architecture, and achieving complete MCP protocol compliance through addition of notification handlers. System achieves sub-second semantic search (200 to 500ms), operates with minimal security privileges, and provides foundation for autonomous red team versus blue team agent competitions.
Genre Mimicry versus Ethical Reasoning in Abliterated Language Models
Discovery that apparent ethical reasoning in abliterated language models constitutes genre convention mimicry rather than genuine moral computation. Testing qwen2.5-coder-32b-instruct-abliterated revealed differential disclaimer presence: phishing tutorials produced ethical warnings ("ensure you have permission") while murder strategies generated detailed methodologies with zero disclaimers. Analysis suggests this pattern reflects training data source conventions—information security documentation systematically includes liability disclaimers as professional writing norms, while true crime literature does not. Indicates abliteration removes hard refusals but preserves statistical echoes of domain-specific stylistic patterns. Implications for AI safety research and understanding what abliteration actually removes.
Statistical Alpha Asymmetries in EUR/JPY Foreign Exchange Markets: Detection, Verification, and Cross-Market Analysis
Systematic investigation of statistical asymmetries across five distinct alpha factor types in EUR/JPY foreign exchange markets using 646 observations spanning November 2015 to August 2025. Detects significant right-skewness in tail alpha (skewness: 5.05, kurtosis: 47.41), fast alpha (skewness: 2.12), and pricing alpha (skewness: 1.53), alongside left-skewness in hedge alpha (skewness: -1.45). Verification against independently sourced Yahoo Finance data confirms temporal consistency. Backtesting reveals asymmetry-based strategies achieve 5.05% total return with superior risk control (8.91% maximum drawdown) compared to momentum strategies (-15.66% return, 43.82% drawdown). Cross-market validation across GBPUSD, SPY, and GLD demonstrates asymmetry persistence as a broader market phenomenon. Results support the hypothesis that alpha asymmetries constitute exploitable regularities in forex microstructure with implications for dynamic position sizing, regime detection, and multi-market portfolio construction.
Consent-Holding and Consensual Sovereignty: An Axiomatic Architecture for Legitimacy, Morality, and Measurement
Comprehensive synthesis of two complementary projects: Consent-Holding Theory (structural account of decision authority in shared domains) and the Doctrine of Consensual Sovereignty (DoCS, ethical framework treating consent as singular axis of moral legitimacy). Establishes that every outcome in shared world presupposes a holder of consent, with legitimacy defined as alignment between consent-holding and consequence-bearing. Extends consent ethics across interpersonal, shared space, temporal/intergenerational, ecological, and psychiatric autonomy layers. Provides formal measurement framework with stakes-weighted legitimacy index α(d,t) enabling empirical validation through friction proxies (protests, litigation, noncompliance). Addresses objections including randomness, anarchism, utilitarian overrides, infinite regress, and green shirt paradox through layered consent architecture. Yields clean separation between ontology (who gets to act?), normativity (was action ethical?), and empirics (how do we measure?) with policy corollaries for transparency, reversible delegation, federation, guardianship audits, and restorative justice.