
Insult AI is an adversarial reasoning engine for live web data. It uses Bright Data to fetch sources, build a plan, pressure-test claims, expose weak narratives, and ground every burn in receipts. The agent declares its plan before it acts, and receipts are tiered by provenance — fetched fulltext, news search, or company source — so a judge can see what was read versus glimpsed. When the target becomes personal, the system pivots into clinical guardrails: it attacks the pattern, not the person, and switches to safety resources when crisis language appears. PlanGuard vetoes identity-targeting steps before any scrape runs, and a tone selector plus crisis fallback keep the same engine safe across three personas — roast, brief, and clinical — that share one Bright Data pipeline and one safety floor. Live web data. Adversarial reasoning. Receipts. Guardrails. Prove me wrong.
31 May 2026