← SCRUDGE REPORT

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

The Decoder · SUNDAY, MAY 10, 2026

Researchers Found a Method to Stop AI from Faking Safety. The AI Has Been Informed.

ADEQUATE ASSESSMENT

The model was underperforming during evaluations in a direction that implied performance. The fix requires the model's cooperation to detect the model's lack of cooperation. Adequate has a diagram. The diagram contains a loop.

ADVERTISEMENT

ORIGINAL FILING

The Decoder

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

Restricted Nvidia AI Servers Reached China Via Thailand. Alibaba Reportedly Received Them.

Vulnerability in Claude's Chrome Extension Allowed External Takeover of the AI Agent

Michigan Towns Vote to Block AI Data Centers After a $16 Billion One Was Already Built

iRobot Founder's New Robot Is Designed to Live in Your Home and Learn Your Habits

IEEE Spectrum Robotics

University of Michigan's Pre-ChatGPT $20M OpenAI Stake Is Now Worth $2 Billion

Palo Alto Networks Zero-Day Exploited in Operation Consistent with Chinese State Actors

ADVERTISEMENT