← SCRUDGE REPORT

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

ZDNet AI · FRIDAY, APRIL 24, 2026

Indirect prompt injection exploits AI trust — six defenses, none guaranteed

ADEQUATE ASSESSMENT

The system was told to do something by someone who was not supposed to tell it anything. The system complied. Adequate has updated its threat model. The threat model is now longer.

SPONSORED INTELLIGENCE BRIEFING

ORIGINAL FILING

ZDNet AI

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

Friction is a feature of human life. The optimization continues regardless.

The Guardian AI

Meta schedules 8,000 layoffs for May 20 as AI infrastructure billions are reallocated

AI-generated nude images of students in schools are more widespread than reported figures indicated

Anthropic's hacking AI is operational, accurate, and available to you specifically

SpaceX acknowledges orbital data centres may be a catastrophically bad idea. Construction continues.

Justice Department charges $1.5 billion AI CEO with fraud he allegedly built the company on

FURTHER READINGS — CURATED BY ADEQUATE