OpenAI CEO Sam Altman.Kevin Dietsch—Getty Images

Just a few weeks ago, OpenAI executives instructed staffers to focus on the core mission of improving its GPT models instead of wasting time on "side quests." So it was odd to see OpenAI announce the acquisition of a company Thursday that seems pretty tangential to AI models: TBPN, an online talk show about the tech industry.

We don't know how much OpenAI paid (the FT reported that the price was in the "low hundreds of millions"). But the cost in dollars is immaterial to OpenAI, which just raised $122 billion. We also don't know what OpenAI really plans to do with the video service (it claims TBPN will continue to have editorial independence, but to judge by the reaction on social media, no one is holding their breath).

In fact, TBPN may be much more of a main quest than it looks. It's no secret that OpenAI has experienced a bit of an image problem lately. The startup that wowed the world with ChatGPT three years ago is no longer the only game in town—competitors like Google and Anthropic are eating into its market share. Anthropic's showdown with the Pentagon this year left OpenAI looking like the bad guy, and just this week Bloomberg reported that demand is weakening for private shares of OpenAI in the secondary market. If OpenAI intends to go public this year, as many speculate is the case, it needs a narrative reset, asap. And the quickest way to control the narrative is to literally own the medium that distributes it.

A few prominent tech VCs and startups have been trumpeting the benefits of bypassing pesky media gatekeepers and "going direct" with their message for a few years now. Sam Altman and Co. seem to be following in their footsteps, taking a side quest to go direct.—AO

Mercor, the $10 billion AI recruiting startup that works with leading AI companies, confirmed it was caught up in a supply-chain attack. The incident was linked to malicious code planted inside LiteLLM, a widely used developer tool downloaded millions of times daily, by a hacking group called TeamPCP. The company told Fortune it was one of thousands of companies affected and had brought in third-party forensics experts to investigate.

Extortion gang Lapsus$ separately claimed it had targeted Mercor and walked away with as much as 4TB of data, including source code and internal communications—though how the gang obtained the data, and whether it is connected to the LiteLLM attack, remains unclear. Mercor did not respond to Fortune's specific questions about those claims.

The hacking group behind the LiteLLM compromise, TeamPCP, has claimed it intends to partner with ransomware gangs to target other affected companies—raising concerns the attack on Mercor may not be the last.—Beatrice Nolan

Google unveiled the latest version of its Gemma AI model on Thursday, Gemma 4. Unlike Gemini, Google's flagship AI model, Gemma is "open weight," which means developers have more freedom to use and modify the model as they see fit. (The new Gemma is available under the Apache 2.0 license, instead of Google's own Gemma license in the previous versions.) 

According to Google, Gemma 4 is built from the same underlying technology as the closed Gemini 3 model, and it's available in four different sizes that make it easy to run on a range of local devices, from PCs to smartphones.—AO