Publications, working research programs, and paid research products on adversarial evaluation, governance lag, institutional readiness, AI safety, and technology geopolitics.
AI capability, institutional oversight, and geopolitical competition move at different speeds. Syntony's research studies the gaps between them, then turns useful findings into client-facing tools, briefings, and advisory packages.
Red-Agent separates model behavior from execution brittleness, then preserves traces, tables, and reports that can be reviewed later.
The public library is small by design. The categories below show where Syntony is building reusable analysis, tools, and client-ready research products.
Campaign design, operational reachability, trace preservation, scoring, and reporting standards for LLM and agent testing.
AI safetyReproduction, learned approval, failure taxonomies, and the problem of making safety evidence usable outside toy settings.
InstitutionsWhere deployment speed outruns oversight, decision rights, measurement, escalation, and control updates.
GeopoliticsCompute, chips, export controls, standards bodies, energy constraints, procurement, and sovereign AI investment.
SystemsCLDs, hazard-to-harm pathways, leverage points, and systems views of AI risk across organizations and markets.
ProductsPaid research packages, diagnostics, watchlists, briefings, and licensed tools built from the research agenda.
Only real Syntony-authored reports and papers are listed here. Each record links to a publication page, source record, or PDF.
Operational reachability accounting, error separation, and artifact-complete reporting across 11 models and 20 probes.
Operational reachability accounting, artifact-complete reporting, and reproducible multi-provider red-team evaluation.
Cross-sector analysis of institutional readiness for AI-enabled systems across defence, energy, finance, and biosurveillance.
Reproduction, learned approval, and design implications for reward-hacking mitigation.
These are the research lanes behind Syntony's advisory work, licensed tools, and paid research products. They are not presented as finished papers unless a publication is linked.
How should red-team campaigns preserve enough context to support governance decisions later?
How should mitigation claims be tested when reproduction is incomplete and learned approval changes behavior?
Where do formal frameworks fail because institutions cannot measure, manage, or escalate after deployment?
Which bottlenecks turn technology adoption into geopolitical exposure or strategic leverage?
How do capability pressure, incidents, regulation, public trust, safety investment, and deployment incentives reinforce or dampen one another?
How should public evidence, proprietary client context, and decision frameworks be packaged for leaders who need to act?
The through-line is practical: what evidence should change a decision, who owns that decision, and what has to be recorded?
Campaign design, trace preservation, evaluator scoring, and escalation thresholds.
Decision rights, review cadence, operational measurement, and action authority.
Compute access, semiconductor capacity, standards bodies, energy constraints, and export controls.
Reproduction, failure taxonomies, learned approval, and deployment-relevant evidence.
Syntony uses research to shape advisory engagements, licensed tools, board briefings, CLD maps, and paid research products. Public work shows the direction; client work applies it to a specific AI risk ecosystem.