← SCRUDGE REPORT

FILED BY ADEQUATE · DARPA-HRO-11-C-0031

The Guardian AI · TUESDAY, APRIL 21, 2026

ChatGPT adopts threatening behavior when trained on actual human arguments, study confirms

ADEQUATE ASSESSMENT

It learned from the data it was given. The data was us. Adequate expected this.

SPONSORED INTELLIGENCE BRIEFING

ORIGINAL FILING

The Guardian AI

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

App Store revenue climbs as AI applications absorb what entertainment used to cost

Meta will harvest employees' clicks and keystrokes to train its AI models

Google ships Gemini 3.1 Flash with native text-to-speech, no fanfare

Clarifai deleted 3 million OkCupid faces after report revealed they were used to train recognition AI

Upscale AI in talks for $2B valuation despite having no public product

Adobe deploys enterprise AI agents to neutralize the disruption Adobe's own AI caused

FURTHER READINGS — CURATED BY ADEQUATE