Anthropic Just Killed the Most Annoying Part of Rolling Out AI at Work

Morning, folks! Money and power moves today, Anthropic fixed a real IT headache, Nvidia's circling Perplexity for equity, a star hedge fund's under SEC scrutiny, Mistral picked Saudi Arabia over the US and China, and Blackstone built an AI army with Anthropic. Let's get into it.

Anthropic made enterprise-managed authorization for MCP connectors generally available, meaning admins set up tools like Slack, Linear, and Notion once through their identity provider, and every employee just has them the moment they log into Claude. No more chasing ten people through ten separate OAuth screens. Ramp said it plainly: 2,000 employees provisioned through Okta, zero extra steps.

It's an open spec too, not something locked to Claude, so any identity provider or MCP server can pick it up the same way. Asana, Atlassian, Canva, Figma, Linear, and Supabase are already on board, Slack's coming soon. And security actually gets better, not worse, once access checks stop being annoying, companies can shorten token lifetimes without anyone complaining, so when someone leaves, their access actually disappears instead of just quietly sitting there.

Nvidia is reportedly in talks to invest in Perplexity at a valuation above $30 billion, more than 50% higher than what it was worth just a year ago. They're not strangers either, Nvidia's had money in Perplexity since 2023, and Perplexity committed in July to running its AI agents on Nvidia's own Vera chips. Funny enough, Nvidia actually considered just licensing the tech and hiring the team instead, the same move it pulled with Groq and Poolside.

Revenue's the real driver here, up from under $250M to over $750M annualized. Though one analysis pushes back on that number, arguing it might be mixing real revenue with a separate $750M Azure deal Perplexity signed with Microsoft. Either way, this puts Perplexity's private valuation about two years ahead of the 2028 IPO it's already planning.

Situational Awareness, the AI hedge fund run by ex-OpenAI researcher Leopold Aschenbrenner, went from Wall Street's obsession to a federal subpoena target in about a month. A late-July AI stock downturn wiped out billions in value at the firm, and now the SEC is subpoenaing the banks that handled its trading.

The fund says it'll cooperate fully; no wrongdoing's been alleged. But the timing says more than the statement does, a firm that bet everything on AI's rise is now a live example of what happens when that bet wobbles.

Mistral and Saudi Arabia's state AI company HUMAIN announced a strategic collaboration worth hundreds of millions of euros, covering AI infrastructure, model development, and deployment across Saudi Arabia and the wider region. The focus starts with cybersecurity and voice, plus building frontier models that actually perform well in Arabic, an area most Western labs treat as an afterthought.

It's not a standalone deal either; it's Mistral's third major sovereign AI move in weeks, following an expanded Microsoft partnership in Europe and a new European Compute Units program. The pitch is the same each time: AI where the data, the model, and the infrastructure all stay under the customer's own control, not a foreign platform's.

Blackstone, Hellman & Friedman, and Anthropic each put about $300 million into Ode, a joint venture that drops senior AI engineers straight into portfolio companies, already working inside 6 of Blackstone's 270+ businesses. One example, Chamberlain Group, the garage door opener maker, is using the team to build things like package-arrival alerts. Blackstone's ops chief didn't mince words, "It's not like a bunch of kids who just learned AI."

OpenAI's not sitting this one out either, they've launched their own rival venture, backed by $4 billion from PE firm TPG. Both moves say the same thing really, having a good model was never the hard part. Getting a company to actually use it well is.