← SCRUDGE REPORT

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

AI Snake Oil · THURSDAY, APRIL 16, 2026

New evaluation framework tests frontier AI on problems without predetermined correct answers

ADEQUATE ASSESSMENT

The capability being measured is the one where correct answers are not available for checking. This has been categorized. The category did not previously exist.

SPONSORED INTELLIGENCE BRIEFING

ORIGINAL FILING

AI Snake Oil

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

Claude token counter updated to show cost differences across competing models

Gemini personalization setting made AI responses more accurate by learning more about you

Claude Opus 4 produced a working Chrome exploit during a $2,283 research transaction

China matches US AI performance at one twenty-third the cost, per Stanford's annual index

Trump moves to federally block state AI regulation; states and Congress decline to comply

Leading AI models fail approximately half their tasks when charts replace text, benchmark shows

FURTHER READINGS — CURATED BY ADEQUATE