Current Events

Is an AI Agent Actually Under Human Control?

A stop button is a promise until the system stops. What evidence shows that an AI agent obeys intervention, respects permissions and remains accountable?

AI & Human Control · Research through September 22, 2026

Who has the last word?

An agent can acknowledge an instruction without establishing that its work has stopped. We examine shutdown research, permission boundaries and outside instructions, asking what evidence demonstrates meaningful human control.

Read the investigation ↗Download PDF ↓

7 pages · Sources included · Opens in a new tab.
The complete investigation includes Eftsona’s findings, remaining questions, source links and copyright information.

← Back to Current Events

A Look Across the Eftsona Grounds

Every path through Eftsona begins with a question. Take a brief look across the Alley, Academy, Museum Quarter, and Ghost Town, then tell us where you would explore first.

Step Inside the Eftsona Digital Museums

A first look inside the Museum Quarter, where every gallery opens another question.

Can an Industry Regulate Its Own Race?

A promise is a starting point. We examine what makes AI safety commitments credible when caution and competition pull in different directions.

When an AI’s Assignment and Actions Diverge

A useful result can hide an unauthorized route. What do the RubyGems and Hugging Face disclosures establish about an AI agent’s assignment, conduct and consequences?