SuperIntelligence - Robotics - Safety & Alignment

About the Journal

We are entering the era of artificial general intelligence (AGI) and foresee beyond it to superintelligence (SI). Assuring safe interaction between AGI/SI and humans is increasingly urgent and important. Ideally, safe interaction means provably safe, i.e. via mathematical proofs, the laws of physics, or ''guaranteed safe' via modeling and simulation. We focus on ’high-risk’ AGI/SI. Realistically, we can estimate AGI/SI timelines but must implement the state-of-the-art safety tech at all times.

SuperIntelligence™ (SI) is open-access. We continuously publish key developments in AGI/SI and AGI/SI safety: theory, technology, and governance. Submissions of articles, expert commentary, editorials, letters, and book reviews are welcome. There are no article processing charges and we emphasize public commentary over traditional anonymous peer review. Authors retain copyright, typically under CC-Attribution-NoDerivs 4.0, which permits further distribution of their work.

Announcements

Dario Amodei, The Adolescence of Technology: Confronting and Overcoming the Risks of Powerful AI

2026-05-05

Anthropic CEO Dario Amodei envisions a 'country of geniuses in a datacenter' with these five enabling properties:1) smarter than top humans across most domains, 2) able to act autonomously over long horizons, 3) use digital tools, 4) coordinate many copies, and 5) operate at much higher speed than humans. You could use these as a checlist to develop such a system, as Anthropic probably does. #5 is true in some domains (real-time learning being a counter-example), #3 is well in progress, and #1, #2, and #4 are still significant obstacles. He does not mention, e.g., robust generalization, abstraction, and world models. The essay discusses risks, governance, and economic implications. The essay’s overall thesis is: AI’s upside remains enormous, but humanity must treat the next few years as a civilizational test requiring technical alignment work, pragmatic regulation, geopolitical realism, economic adaptation, and moral seriousness.

Current Issue

Vol. 3 No. 2 (2026): Cybersecurity and Artificial General Intelligence (In progress)

Co-founding editor Steve Omohundro and others, notably ex-Google CEO Eric Schmidt, predict that artificial general intelligence (AGI) will find and exploit every possible hardware and software vulnerability, whether through its own instrumental goals or in the hands of bad actors. Some, including Senior Editor-at-Large Gil Syswerda, predict an escalation of cyberattacks throughout 2026-2027, culminating in infrastructure outages. Imagine no internet, no access to your money, no operating gas or electric charging stations…

In this issue we look at the state-of-the-art of cybersecurity, cyber offense and cyber defense.

Published: 2026-05-08