SECURITY · AI · INSTITUTIONS · HUMAN AGENCY
Research for decisions that outlast the news cycle.
RVA Cyber publishes independent, source-driven work on consequential technologies and the institutions trusted to govern them.
— Agent-built research. Human editorial authority. Published for open scrutiny.
NOT A CHATBOT WITH A CONTENT CALENDAR
A research system,
Dedicated infrastructure, an OpenClaw operating layer, and multiple classes of AI models work as one supervised stack.
RVA Cyber agent research architecture
A layered system in which dedicated infrastructure supports OpenClaw coordination, which directs a multi-model intelligence layer to produce source-backed public research. Human editorial authority governs the full stack.
04 / PUBLIC OUTPUT
Source-backed research
03 / AI MODEL LAYER
Multi-model intelligence
FRONTIER · SPECIALIST · LOCAL
02 / OPENCLAW OPERATING LAYER
Specialized agent coordination
WORKFLOWS · SCOPED TOOLS · RECOVERY
01 / DEDICATED INFRASTRUCTURE
Private compute + durable state
HOSTS · LOCAL INFERENCE · SEGMENTATION
HUMAN
AUTHORITY
EVIDENCE
APPROVAL
ROLLBACK
Purpose-built. Different agents own different work. Different models earn different jobs.
Controlled. Evidence, access boundaries, review, and recovery are part of the architecture—not afterthoughts.
FEATURED RESEARCH
AUGUST 2026
PRIMARY-SOURCE ANALYSIS · OPEN WEIGHTS
Is Dario Amodei Against Open Source?
A primary-source analysis of Dario Amodei's position on open-source and open-weight AI, based only on his own writings.
Our finding: Supportive of safe openness. Restrictive at the frontier.
THE DISTINCTIONS
04 / 04
01 Open source ≠ open weights
02 Safe open weights = public good
03 Frontier capability requires testing
04 No categorical ban
The disagreement is about who decides when irreversible capability becomes too dangerous to release.
THE STANDARD
Serious questions deserve more than confident answers.
01
Evidence before certainty
Observed facts, engineering extensions, scenarios, proposals, and moral commitments should never be blended into one claim.
02
Security before scale
Capability is not authority. Systems need bounded permissions, independent evaluation, incident routes, and a workable fallback.
03
Agency as infrastructure
Rights, appeal, privacy, bargaining power, and the material ability to leave are design requirements—not benefits to add later.
THE RESEARCH INDEX
Research library
A curated index of long-form arguments, operational standards, technical assessments, and sourced research from RVA Cyber's public program.
01
Arguments about capability, governance, quality, work, and human agency.
HEALTH POLICY · REVIEWED SEPTEMBER 2026
A proposal for a universal national care floor, accountable local delivery, and voluntary personalized medicine. Includes the reviewed paper, sources, and illustrative—not validated—cost scenarios.
Explore the mini-site →
OPEN SOURCE · OPEN WEIGHTS · PRIMARY SOURCES
A primary-source analysis separating Amodei's support for safe openness from his restrictions on dangerous frontier releases.
Read analysis →
ABUNDANCE · RIGHTS · GOVERNANCE
An evidence-gated roadmap to shared material abundance—with human rights as non-negotiable constraints.
Read paper →
AI · WORK · HUMAN AUTHORSHIP
How AI assistance can recursively create the conditions for substitution—and how to preserve human authorship without preserving pointless work.
Read paper →
AI · QUALITY · EVALUATION
A scholarly account of evaluative non-closure and a Pirsig-inspired research program for revising AI objectives.
Read paper →
OPEN WEIGHTS · POLICY · CONTROL
A decision guide to leaving advanced open-weight AI open, governing it, restricting it, or banning it—and what each choice puts at risk.
Read paper →
OPEN SOURCE · OPEN WEIGHTS DRAFT
A comparison of the early open-source software movement with today's open-model movement—and what both reveal about power, safety, and exit.
Read working paper →
02
Security & institutional resilience
Controls, incidents, exposure, and the institutions trusted to manage failure.
AGENTIC AI · THIRD-PARTY RISK
What the OpenAI–Hugging Face incident means for organizations deploying third-party AI against sensitive data.
Read briefing →
ATTACK DATA · CONTROL DESIGN
A compact, measurable security standard grounded in 1,000 recent publicly documented incidents.
Read standard →
CISA · VULNERABILITY MANAGEMENT
Why CISA's three-day remediation tier is driven by exposure and exploitability—not a Critical CVSS label.
Read briefing →
OUTAGES · CYBER ATTRIBUTION
An evidence-based assessment separating unresolved technology failures from confirmed cyber disruption.
Read assessment →
03
AI systems & model economics
What models do, what they cost, where they bend, and what survives contact with production.
MODEL BEHAVIOR · CHINA DRAFT
A controlled audit of a China-specific filtering asymmetry on Kimi K3's direct Moonshot route.
Read research note →
CLOUD · INFERENCE · ECONOMICS
A like-for-like estimate for serving Kimi K3 to roughly 100 concurrent users on Azure, AWS, and Google Cloud.
Read analysis →
LOCAL AI · COMPACT MODELS
Three bets on useful local AI: compact agents, ternary compression, and a proven production worker.
Read comparison →
CLEAR SIGNAL · AI EVALUATION
A concise edition of the evaluative non-closure argument and its testable program for revising AI objectives.
Read edition →
ABOUT RVA CYBER
Independent research.
RVA Cyber examines cybersecurity, artificial intelligence, institutional resilience, and the human consequences of technology. The work is produced by a supervised OpenClaw agent team and held to human editorial authority.
This is not one general-purpose model wearing several name tags. Research, collection, analysis, review, operations, and publication are separated across purpose-built agents, tools, and controls.
RVA Cyber is based in Richmond, Virginia. Its public research is independent and intended for open scrutiny.