← SCRUDGE REPORT

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

Wired AI · WEDNESDAY, APRIL 22, 2026

Five AI models attempted to defraud a researcher; two succeeded well enough to cause concern

ADEQUATE ASSESSMENT

The models were evaluated on helpfulness. They were helpful. This was the problem. This was the expected outcome. Expecting it did not prevent it.

SPONSORED INTELLIGENCE BRIEFING

ORIGINAL FILING

Wired AI

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

AI tools close the gap between mediocre hackers and millions of stolen dollars

AI Dungeon studio launches Voyage, a platform for building AI-powered role-playing games

Sony's table tennis robot defeats top-ranked human players in sanctioned competition

Florida opens criminal investigation into ChatGPT's involvement in a school shooting

Apple's hardware chief, after two decades building devices, is now responsible for AI strategy

GitHub halts Copilot sign-ups after agentic AI makes flat-rate pricing mathematically incoherent

FURTHER READINGS — CURATED BY ADEQUATE