GeraWitness Research
The human-in-the-loop safety layer for agent-initiated real- world actions. Humans reviewing the commit step on high- risk transactions.Coming soon — join the waitlist
What GeraWitness Will Be: The Safety Layer for Agent Actions
A human-in-the-loop escalation layer that sits between agents and irreversible real-world actions — always-available humans reviewing the commit step on high-risk transactions.
2026-04-21 · 6 min read
ThesisWhy Agent Safety Needs Humans in the Loop
Full autonomy is the wrong target for commercial agents. A calibrated human-review layer is cheaper than the class of mistakes it prevents.
2026-04-21 · 8 min read
TechnicalA Human-in-Loop Escalation Design for Agent Safety
Technical sketch of how GeraWitness routes agent actions by risk tier — automatic, sampled, mandatory human review, and hard stop.
2026-04-21 · 10 min read
ResearchOpen Questions: Human Oversight at Scale
Reviewer burnout, cultural calibration, training the reviewer pool, conflict of interest, cost per review. The hard parts of human-in-loop safety.
2026-04-21 · 7 min read
PositioningGeraWitness vs. Constitutional AI vs. OpenAI Moderation: Layers, Not Rivals
RLHF and constitutional training shape agent behaviour at inference. Moderation APIs catch content. Human-in-the-loop catches the things neither can. All three belong in the stack.
2026-04-21 · 8 min read
TechnicalProtocol Deep-Dive: Risk Tiers, Review SLA, Reviewer Accountability
Four risk tiers, per-tier SLAs, credentialed reviewers, signed decisions — the mechanics of human-in-the-loop review.
2026-04-21 · 12 min read
VisionA World With Human-in-the-Loop Agent Safety in 2030
The review profession, the mistakes that stop happening, the dispute volume that falls, and the new workforce that emerges.
2026-04-21 · 7 min read
PositioningGeraWitness vs. Scale Surge vs. In-house Trust & Safety: Adjacent Approaches Compared
Fair comparison with Scale Surge data labelling, platform T&S teams, and open-source moderation toolchains.
2026-04-21 · 9 min read
ResearchFAQ: Common Objections to Human-in-the-Loop Review, Answered
Latency, cost, reviewer bias, scale, labour conditions, why-not-AI — twelve objections answered directly.
2026-04-21 · 6 min read
RoadmapRoadmap Preview: GeraWitness v0 to v1
Risk tiers, reviewer pool, third-party providers, regulator integration — phases and open questions.
2026-04-21 · 4 min read