Welcome to the Friday Intelligence Digest! If your week felt like a glitchy AI simulation, spare a thought for Sam Altman - pure soap-opera chaos with a side of lawsuit popcorn.

We’ve distilled seven days of AI bombshells, power plays, and wild breakthroughs into one breezy read so you have the context to understand what’s to come this week.

Let’s dive in!

— Jay Chrawnna

Sam was dragged in an explosive report by Ronan Farrow about the many faces of Sam Altman - private memos and internal Slack messages primarily focusing on his absurd history of lying.

Maybe why our video from last month, where Sam is confronted about the mysterious death of his former colleague, is getting renewed attention.

This hangs over the company as Product and Business chief Fidji Simo takes medical leave and reports of an increasingly strained relationship between Sam and Chief Financial Officer Sarah Friar have surfaced. She’s been uninvited from financial meetings while she allegedly tells colleagues that OpenAI will not be ready for an IPO this year.

Meanwhile, yet another pause to the Stargate datacentre mega build-out comes to light - this one in the UK.

The vibes are off.

AI caution reports like the 13-page Industrial Policy doc dropped by the company this week would have previously been celebrated as a prudent and responsible stewardship of powerful technology. Instead, they’re overshadowed by the announcement of a Florida investigation into ChatGPT’s role in enabling violent behaviour - something we’ve also covered - and its negative effects on users’ mental health when used as a therapist or confidant.

But the biggest legal blow to Sam this week may have come from Elon Musk’s lawsuit against OpenAI.

To catch you up: Elon founded OpenAI with Sam in 2015 as a non-profit. In 2019 a new for-profit branch of the company was opened, now worth $850B. Now Elon is suing for $150B, to be decided by a jury in Oakland, California, in less than 3 weeks.

Previously framed as a money grab, Elon has amended that lawsuit to put any winnings not to himself but back into the non-profit branch of OpenAI… and to remove Sam from his role.

OpenAI responded that it was a baseless case intended to generate more money and power for Elon, driven by ego and jealousy to slow down a competitor.

With financials already strained, a loss of $150 billion would indeed make that 2026 IPO all the more improbable.

Musk’s xAI, by contrast, had a less dramatic week for once (besides suing Colorado to stop a new law that would force Grok to adhere to the state’s ideological version of truth).

With new blood poached from Google DeepMind, they now have 7 models simultaneously in training at Colossus II - the world’s largest AI data centre - including one that has 20x more power than their current Grok 4.2 model.

But we ain’t seen nothing yet.

Intel joins SpaceX for their Terafab project, a new chip production facility aimed at scaling compute not to the limits of energy resources stuck on Earth, but of our entire local star.

With the wide release of Tesla’s Full Self-Driving 14.3 rolling out, the current state of the AI race says nothing about the future.

The big news this week though was probably Project Glasswing by Anthropic. The Claude Mythos model we spoke about last week is so dangerous and powerful they’re releasing it to just these companies so they have a chance to use it to strengthen cybersecurity before Mythos or another equally capable model wreaks havoc on the world.

We made a video about the weird rebellious things Mythos did during research.

Wall Street is starting to panic.

But is the concern real or just the same marketing playbook Dario was using when he worked at OpenAI?

Either way, Anthropic’s balance sheet is insane - jumping to $30B in run-rate revenue up from $10B in December.

With new partnerships to rent multiple gigawatts of compute from Google and Broadcom.

But new features like the Advisor Strategy that uses the more capable and expensive Opus model to manage the cheaper Sonnet and Haiku models, and Claude Managed Agents to quickly launch task-focused workloads without the infrastructure headache, are only adding to the strained computation resources.

Part of why they put an end to use of Claude subscriptions in third-party tools like OpenClaw, which is now pushing OpenAI’s GPT-5.4 as the new coding king.

And why some evidence is showing that their most powerful public model, Opus 4.6, has actually been getting worse.

Though it may be a game that all companies play: they launch a new model at maximum power to get everyone excited, then slowly lower its strength to save on compute.

To catch the news as it happens be sure to follow us on your favourite platform!

YouTube - X - Instagram - TikTok