u9up:~$ whoami
U9UP /9up/ verb
An independent AI security community from Malaysia. We publishes hands-on research, raising awareness, and sharing what we learn.
Deep dives into AI security blind spots, written by people who actually break them.
git clone && get owned: one poisoned supplier reaches everyone downstream. Pickle-based model formats run code on load, and coding agents moved the trust boundary from “you wrote it” to “can the agent tell intent from injection?”
Models, datasets, plugins — the AI stack has a supply chain, but none of the controls. Walking a poisoned MCP server from marketplace upload to backdoor on your host.
Full technical breakdowns of bugs we found and reported, with proof-of-concept and the fix.
Headroom reads where a request should go and who is making it from headers the caller sets, and it checks neither. That’s enough to reach cloud metadata, steal the API key, and read or write another user’s memory.
Opening a marimo notebook can run an attacker’s command with zero clicks, and the same config-injection flaw can leak the operator’s API key. Two CVEs, one root cause.
AI security isn't just breaking things or defending them, it's both. We research across the full spectrum.
Finding the cracks before the bad actors do. We research attack vectors, test guardrails, and publish what we find so the community can learn.
Building resilience into AI systems. We study detection, guardrail architectures, and monitoring strategies that actually work in production.
Workshops, talks, and the occasional makan session.
Whether you're deep into adversarial ML or just getting curious about AI security — pull up a chair. No gatekeeping.
