AI Sector Daily Digest — June 26, 2026

The busiest AI signal today is not one lab announcement. It is the same pattern showing up in models, policy, supply chains, and funding: frontier AI is becoming more regulated, more geopolitical, and more infrastructure-heavy.

Coverage window: June 25 08:00 to June 26 08:00.

The five

1. OpenAI moves GPT-5.6 into a controlled preview

OpenAI will release GPT-5.6 first to selected partners, with customer-by-customer government approval during the preview, according to reporting cited by Reuters; Politico reported the access list will be limited to government-approved organizations. 1 2
The same Reuters story says OpenAI is weighing a later IPO, with advisers presenting a choice between a faster listing at a lower valuation or waiting until 2027 to pursue a $1 trillion target. 1
Source: Reuters and Politico.

2. Z.ai says GLM-5.2 is closing the open-source frontier gap

Chinese AI company Z.ai said it will use domestic listing proceeds to fund its AGI push after GLM-5.2 scored near leading U.S. closed models on public benchmarks. 3
Reuters reported GLM-5.2 has 750 billion total parameters, a 1-million-token context window, fourth place on Artificial Analysis' LLM intelligence leaderboard, and second place on Code Arena's front-end coding leaderboard. 3
Source: Reuters.

3. The EU joins the U.S.-led Pax Silica AI supply-chain pact

The European Commission joined Pax Silica, a U.S.-led initiative for allied-country coordination on AI supply chains, European Commission spokesperson Paula Pinho told Reuters. 4
The pact covers access to energy, critical minerals, high-end manufacturing, AI models, and broader supply-chain security; the Netherlands joined earlier this week. 4
Source: Reuters.

4. Italy's Domyn promises a fully open-source frontier model

Domyn CEO Uljan Sharka told Reuters the company plans to release a fully open-source frontier AI model within a year, built through its EUROPA consortium with Germany's Fraunhofer-Gesellschaft. 5
The planned model will have more than 400 billion parameters, be trained from scratch, and be reproducible so companies and governments can run it on their own infrastructure. 5
Source: Reuters.

5. General Intuition raises $320 million for game-trained agents

General Intuition raised $320 million at a $2.3 billion valuation to train agentic AI on gameplay action data and apply it to simulation, robotics, and embodied systems. 6
TechCrunch reported the round was led by Khosla Ventures, brought total disclosed funding to $454 million, and will mostly go toward compute for the next model version. 6
Source: TechCrunch.

What changed since yesterday

Government review is no longer just a background risk for frontier labs: the OpenAI rollout and Pax Silica story both point to more state involvement in who can access models, chips, energy, and data-center inputs. At the same time, Z.ai, Domyn, and General Intuition show capital and model-building activity moving outside the narrow OpenAI-Anthropic-Google frame.