The Cutoff ReportAI & startup funding, tracked daily Search

OpenAI scrambles academic trust with new science discovery model

The AI lab reveals three new breakthroughs in physics and math while shifting how it collaborates with outside experts.

October 27, 2026•1 min read•Based on reporting by The Verge

Abstract geometric shapes and lines representing scientific discovery

OpenAI has announced three major scientific breakthroughs spanning the fields of physics, computer science, and mathematics. This development follows a previous release of hundreds of manuscripts that the company claimed solved enduring mathematical problems.

To refine and validate these latest findings, the organization has introduced a new collaboration method. This approach involves utilizing external specialists, whom the company refers to as shepherds, to oversee and guide the results.

What it means

This shift toward utilizing outside academic guidance marks a change in how the AI lab handles the verification of its scientific output. By integrating external expertise, the company aims to provide a more structured layer of oversight for its technical discoveries.

Sources

https://www.theverge.com/ai-artificial-intelligence/1019143/openai-eth-kaseya-science-math-physics

This article was written and illustrated with AI assistance from the reporting credited above, and reviewed against our editorial & AI policy. Facts come only from the linked sources. Spotted an error? Tell us.

More in AI

AI/Nov 1

Gemini 3.8 Flash hits 100% benchmark score on 2026-11-01

Gemini 3.8 Flash achieved a 100% benchmark score. Six models returned to service after failed checks, with uptime across hourly checks hitting 70.8%.

AI/Nov 1

Lab Risks: AI Agent Administers Gene Therapy to Live Mouse

A biotech company successfully used an AI agent to identify a specific gene therapy and deliver it into a live mouse's brain. This event demonstrates the potential for autonomous agents to handle physical laboratory tasks.

Abstract vector art showing neural connections and a medical drop.

AI/Oct 31

Gemini 3.8 Flash and GPT-5.6 Terra Hit 100% Benchmark Score (2026-10-31)

Gemini 3.8 Flash and GPT-5.6 Terra achieved 100% on standing benchmarks. However, all tested models across Google and OpenAI showed 0% uptime in 24-hour speed checks.

AI/Oct 30

Gemini 3.8 Flash and GPT-5.6 Terra Hit 100% Benchmark, 2026-10-30

Gemini 3.8 Flash and GPT-5.6 Terra both recorded 100% on the standing benchmark. All measured Gemini models and GPT-5.4/5.6 variants failed uptime checks over the last 24 hours.