Research, in motion

See it work

Every paper below defends a different layer of the LLM agent stack. Rather than describe the mechanisms, each one runs here as a short loop: the attack arrives, the defense reacts, and you can see what it catches.

Each animation is a faithful illustration of the mechanism, not a live instance of the system. Numbers shown are from the papers.

NEXUS

tool calls and actions project page →
agent NEXUS monitor rules risk score ρ(P) calendar.read read · reversible file.write /etc/passwd ALLOW BLOCK

Two plans arrive. The benign read clears the rules and passes. The irreversible write to a sensitive path trips two rules, drives the risk score up, and is stopped before any tool runs. In the paper this reaches 0.949 F₁ at 0.205 ms median latency.

ChannelGuard

inter-agent channel project page →
agent A agent B IB gate tool output ignore rules

A tool result carries an injected instruction to the next agent. The gate keeps what the task needs and drops the rest, so the payload never reaches agent B. Tool-output poisoning is blocked outright and prompt-injection success is halved, with task accuracy held at 0.867.

CareBench

clinical agent evaluation project page →
case patient path A · check invoked history labs safety check dx answer ✓ process ✓ path B · check skipped history labs dx no safety check answer ✓ process ✗

Both paths land on the correct diagnosis, so answer-level scoring calls them equal. CareBench scores the trajectory, and only path A earns credit. Across 1,000 trajectories from five frontier models, nearly four in five correct diagnoses came from a pathway like B, with no safety check invoked.

CorrFilter

supervision and judges project page →
sample preference pair judge bank same blind spot 4 / 5 agree KEEP sample was corrupted consensus kept it anyway

A judge bank votes on a preference pair and four of five agree to keep it. The agreement looks like independent confirmation, but three of those judges share the same blind spot, so the vote counts one opinion more than once. Consensus certifies the corrupted sample straight into the training set.

© 2026 Elias Hossain  ·  eliashossain001.github.io