00 / RED TEAM
ECHOS
Adversarial security research against self-hosted MCP servers and the agent infrastructure around them. Real vulnerabilities, disclosed responsibly. No targets named before disclosure.
About ECHOS→First advisories Q4 2026
Offensive Security · AI AgentsEst. 2023 · Oklahoma US
botzrresearch.
Independent offensive security research: AI agent systems, MCP infrastructure, and the memory they trust. The work happens in public. Findings are the product.
01 / TOOLING
AEGIS
A Rust and WASM sandbox for detonating and constraining untrusted MCP tool calls: an eight-crate policy → capability → sandbox → audit pipeline, MIT, denied by default. Built and open source; cited here as past work, not an active program.
About AEGIS→Built 2026 · archived
02 / BENCHMARK
dreamd
A reproducible benchmark for how AI memory systems handle facts that change, including facts an attacker changes. Built, then archived before the runs went public. No results were published, and none are claimed.
github.com/botzrDev→Archived August 2026
03 / PRIMER
Memory as attack surface
What happens to an AI system's stored beliefs when the facts underneath them change, or are made stale on purpose: what the field has measured so far, and what remains untested.
Read the primer→Background
04 / PRINCIPAL
Uveddi
Austin Green. Independent researcher and systems engineer in Oklahoma. 25 years of systems judgment, applied in public. The work speaks; the person is secondary.
Oklahoma, US