MIRI TGT Blog

This TGT Fellowship project investigates various trends in inference scaling and their impact on future AI governance.

In this blog post, we describe a framework for predicting whether "warning shots" will motivate an effective response, and apply this framework to various candidate warning shots.

This SPAR project investigates interconnect limits, a method of verifying that AI chips are not conducting training.

This podcast discusses AI risks and how humanity should respond, covering some of TGT’s previous work.

This TGT Fellowship project investigates how trusted execution environments (TEEs) could be used for verification in AI governance.

A survey of potential side-channels in data centers and retofittable defenses.

Could covert distributed training undermine international restrictions on AI development?

This blog post discusses deterrence in the context of AI development and the MAIM framework introduced in a recent paper.

This technical blog post presents novel analysis indicating that the rate of catch-up algorithmic progress may be much faster than many previously expected.

Announcing the 2026 MIRI Technical Governance Team Research Fellowship

This post addresses a common objection to our proposed international agreement to halt ASI development: that some countries would avoid joining the agreement and might pursue ASI themselves or weaken the agreement in other ways.

This post explains how our recently proposed international AI agreement could be thought of as similar to a defensive military coalition.

This post addresses a common objection to our proposed international agreement to halt ASI development: that countries would simply cheat by pursuing covert ASI projects.

This post explains our reasoning for the FLOP thresholds in our proposed international AI agreement.

We at the MIRI Technical Governance Team have released a report describing an example international agreement to halt the advancement towards artificial superintelligence. The agreement is centered around limiting the scale of AI training, and restricting certain AI research.

The Artificial Intelligence Risk Evaluation Act is an exciting step toward preventing catastrophic and existential risks from advanced artificial intelligence.

A short exploration of inference prices for large language models using publicly available data.