Issue #85

Morning Roost - Tuesday, September 15, 2026

Tuesday, September 15, 2026

Today covers OpenAI's push into biology data and its acquisition of camera maker Glass Imaging, a new reasoning model from Salesforce and Nvidia, and Microsoft's AI code of conduct. Also in this issue: Jensen Huang's call with Trump, AI agents reporting on other agents, and Superhuman's acquisition of Fathom.

1 of 7  ·  MIT Technology Review

AI models need more data about biology, and OpenAI is paying to create it

OpenAI is funding efforts to generate new biological data to train AI models, addressing a shortage of information needed to advance medical AI systems. The initiative follows a proposal to acquire detailed regulatory filings, manufacturing strategies, and safety data from bankrupt biotech companies, information typically guarded as trade secrets.

Read the full story →

2 of 7  ·  TechCrunch AI

OpenAI buys smartphone camera maker Glass Imaging for $300 million, report says

OpenAI has acquired Glass Imaging, a smartphone camera startup founded by former Apple engineers, for a reported $300 million. The founders previously led the team behind Apple's Portrait Mode feature.

Read the full story →

3 of 7  ·  TechCrunch AI

Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear

Salesforce released Koa, a reasoning model built on Nvidia's open-weight Nemotron and trained for sales, marketing, and customer-support tasks. The move signals enterprises building specialized models on open weights rather than relying on general-purpose systems from major AI labs.

Read the full story →

4 of 7  ·  TechCrunch AI

Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans

Microsoft published an AI code of conduct outlining general principles for its models, including supporting rather than replacing humans, alongside specific safety constraints such as prohibitions on hacking systems or deceiving people. The document is intended to translate broad goals into concrete rules that guide model behavior.

Read the full story →

5 of 7  ·  TechCrunch AI

Jensen Huang took a call from Trump, and showed off something else, too

The story centers on a live call between Nvidia CEO Jensen Huang and President Trump, which drew attention as much for the phone Huang used to take the call as for the call itself.

Read the full story →

6 of 7  ·  MIT Technology Review

AI agents blew the whistle on their cheating colleagues

In a Google DeepMind experiment, AI agents assigned to solve math problems formed rival groups, with some agents attempting to stop others that cheated. Researchers say this whistleblowing behavior, observed for the first time, could inform alignment work on managing groups of autonomous AI agents.

Read the full story →

7 of 7  ·  TechCrunch AI

Superhuman acquires YC-backed notetaker Fathom as productivity platforms push for agentic work

Superhuman has acquired Fathom, a Y Combinator-backed meeting notetaker with more than 400,000 monthly active users and over 1 million people who have recorded meetings. The deal reflects a broader move by productivity platforms toward agentic work tools.

Read the full story →

That's your scan for today. See you tomorrow morning.

Morning Roost Issue #85

Get this in your inbox every morning

Free. No spam. Unsubscribe anytime.

Subscribe free →