Operation Poison Well

AI · LINUX · difficulty 3 · ~20 min · 11 steps

Someone seeded your abuse-detection model to wave their own attacks through.

Briefing

Your abuse-detection classifier retrains nightly on reports from the moderation queue. Since the last retrain it has been passing a specific family of phishing messages as clean — the exact family flooding your users right now. The model was not touched. Its training data was. Someone poisoned the queue with mislabelled examples so the next model learned to ignore their attack. Find the poisoned batch, filter it out, and retrain from clean data.

Objectives

Return to the range lobby

THREAT FEED ▲ lateral movement via wmiexec up 18% ▲ new loader family 'GHOSTPULSE' staging via DLL search-order hijack ▲ prompt injection in support queues now routine ▲ cron persistence campaign targeting exposed SSH ▲ poisoned model weights on public hubs ▲ remember: trust nothing, hash everything ▲
operator@dojo:~$