Brief

Hassabis AGI test chatter points to invention, but primary evidence is still missing

X News summaries on June 13, 2026 say Demis Hassabis discussed invention-style AGI tests, including deriving relativity from historically limited data or creating a game as complex as Go. The same X cluster also points to claims about a Google DeepMind “From AGI to ASI” paper, but the research pack does not include a primary DeepMind report, PDF, blog post, podcast, or transcript. That leaves the evidence strong enough to track a public debate over whether AGI claims should be judged by novelty and discovery, but not strong enough to say DeepMind has formally shifted its evaluation standard.

AI · June 13, 2026

Sources & context

Evidence, open claims, and reader-facing editor notes.

Dek

Demis Hassabis’s proposed invention tests and a new DeepMind AGI-to-superintelligence report make the live question whether AI labs are changing the standards by which capability claims are judged.

Synthesis angle

Investigate whether DeepMind’s public AGI framing shifts institutional incentives from benchmark performance toward harder-to-game tests of novelty, discovery, and autonomous research capability.

Source

Demis Hassabis Sets High Bar for True AGI with Invention Tests

Preferred source plan

Google DeepMind report on AGI and superintelligence paths
Full podcast or transcript featuring Demis Hassabis’s AGI test comments
Google DeepMind official blog or research page for related AGI definitions
Prior AGI definition work by Shane Legg or DeepMind researchers
Independent benchmark or evaluation papers on novelty, discovery, and autonomous scientific reasoning

Reading the dossier...