The brief · 7 July 2026

The AI brief, 2026-07-07

Insilico moves into Phase III with a drug designed by AI. Meta drops Midjourney, Illinois tightens the screws, and Brussels rolls out its plan.

2 min read J / K to navigate

Insilico's AI-Designed Drug Reaches Phase III

Rentosertib treats idiopathic pulmonary fibrosis, a disease that destroys lung tissue with no truly effective treatment available today. This isn't AI just screening existing molecules: it identified the target and designed the drug itself, start to finish. Two years after their partnership with Takeda on other targets, Insilico is now pushing its own candidates all the way through.

practitioners › Rentosertib (formerly ISM001-055) targets the TNIK protein, Phase III launched July 7, 2026, Insilico Medicine.

Insilico Medicine →

Meta ditches Midjourney, ships its own image generator

Meta is cutting third-party tools loose and launching Muse Image, its first in-house model, built by Meta Superintelligence Labs. If you're generating images in Meta AI, Instagram, or WhatsApp, you're already using it without ever choosing to. The real story isn't image quality, it's Meta taking full control of the pipeline.

practitioners › Muse Image has been live since July 7 in Meta AI, Instagram Stories (US), and WhatsApp (select countries), with Midjourney being phased out.

Meta →

Illinois Forces AI Giants to Put Their Risks in Writing

Pritzker signed a law requiring major AI developers to get their safety plans audited by an independent third party. They'll also have to make their risk assessments public. It won't change how you use ChatGPT tomorrow morning, but it adds Illinois to the growing list of states refusing to wait on Washington.

practitioners › The Artificial Intelligence Safety Measures Act, signed July 7, 2026, mirrors the model already in place in California and New York.

source →

Brussels rolls out its AI and cybersecurity action plan

The European Commission wants to test advanced AI models for security flaws before they cause damage, not after. In practice, this pushes providers to document their models' cyber risks and coordinate response with member states. Whether companies get ahead of this before it becomes mandatory remains an open question.

practitioners › The plan was published on 7 July 2026 by the European Commission, including a model evaluation and testing track run in coordination with member states and industry.

Commission européenne →