HALTLINE
Independent AI Oversight & Human Control

Autonomous AI needs an independent line it cannot cross.

Haltline develops independent systems for monitoring, evaluating and constraining increasingly autonomous artificial intelligence while preserving meaningful human authority.

Public-interest initiative Solve & Serve Inc. haltline.org
Oversight loopHuman authority retained
AI SYSTEM HALTLINE MONITOR HUMAN AUTHORITY OBSERVE ESCALATE AUTHORIZE CONSTRAIN HALT LINE — INDEPENDENT OF THE SYSTEM IT SUPERVISES
Three areas of work

Monitor. Control. Report.

Oversight only counts when it is independent of the system it supervises. Each area of Haltline is built to stand apart from the AI it watches, and to keep a human able to say stop.

01

Monitor

AI systems designed to supervise other AI systems — observing behavior continuously and flagging deviation before it compounds.

AI supervising AI
02

Control

Independent safeguards and enforcement infrastructure — the ability to constrain, pause or halt a system, held outside that system's reach.

Independent safeguards
03

Report

A public channel for reporting abnormal or concerning AI behavior — open to researchers, operators and the public.

Public channel
How Haltline works

Monitoring, evaluation and constraint — with a person at the end of the line.

Step 01

Observe

Independent monitors watch an AI system's actions and outputs from outside its own control.

Step 02

Evaluate

Behavior is measured against defined limits, and anything abnormal is escalated to human reviewers.

Step 03

Constrain

Safeguards act on human authorization — limiting, pausing or halting the system. The final decision stays with people.

Meaningful human authority is not a feature to be added later. It is the line the system is built around.
Haltline founding principle

Independent

Oversight that does not depend on the system it oversees.

Public-interest

Run as a nonprofit initiative, accountable to the public.

Open

Methods and findings shared so others can verify them.

Human-final

Every halt is authorized by a person, never only by a machine.

Report abnormal AI behavior

Seen an AI system act outside its limits?

Haltline accepts reports of abnormal, unsafe or uncontrolled AI behavior from anyone. Reports are reviewed by people. The full reporting portal is in development; email is open now.

report@haltline.org Reporting portal — coming soon
Confidential · Reviewed by people