The AI industry just can’t help itself. This week brought a perfect storm of absurdity: companies racing to go public so fast that San Francisco realtors are now accepting Anthropic stock instead of cash for houses. Meanwhile, Donald Trump signed an executive order on AI safety that’s somehow both overhyped and underwhelming, hackers figured out they could trick Instagram’s AI chatbot into handing over high-profile accounts, and a DOGE whistleblower is suing Elon Musk claiming the billionaire’s posts led to someone cutting his car brakes.
Let’s unpack this mess, shall we?
The IPO Madness Has Officially Begun
Anthropic officially filed its paperwork to go public this week, and the timing couldn’t be more emblematic of where we are in this AI bubble. The company’s current valuation sits at a mind-boggling $965 billion, which would make it one of the largest IPOs in history if it pulls it off.
But Anthropic isn’t alone in this race. xAI (SpaceX’s AI venture) has already filed its paperwork, rumored for a June IPO. OpenAI, true to form, refused to confirm, deny, or “speculate, pontificate, or elaborate” when asked directly. Classic OpenAI.
Here’s where it gets weird. Multiple real estate listings in San Francisco have started accepting stock in Anthropic or OpenAI as paymentinstead of cold hard cash. One listing specifically requested Anthropic stock only, because—get this—the seller apparently likes Anthropic’s products better. These aren’t public companies yet, so technically these transactions would require board approval and probably violate all sorts of securities rules. But agents are apparently taking the inquiries seriously.
It’s either a brilliant marketing stunt or the clearest sign yet that we are deep, deep into AI mania. Probably both.
Trump’s AI Executive Order: All Smoke, No Fire?
The Trump administration announced this week that AI companies would voluntarily give the federal government 30 days of advance notice before releasing their most advanced models. That’s down from the original 90-day proposal that Trump himself axed a few weeks ago because, in his words, it “could have been a blocker” to America’s AI lead over China.
The key word here is “voluntary.” This isn’t regulation. It’s barely even a suggestion. The order has no enforcement mechanism, no penalties for non-compliance, and the whole thing reads more like a favor the government is asking rather than a rule they’re imposing.
AI companies, predictably, are absolutely fine with this. They get to look cooperative while essentially maintaining the status quo. As Zoë Schiffer pointed out on the podcast, this order basically lets companies give the government “a taste” of their models—maybe they’ll get some feedback, maybe they’ll land a government contract down the line.
The administration is already using this as diplomatic leverage, with Treasury Secretary Scott Bessent exploring similar frameworks with China. But let’s be honest: 30 days isn’t enough time to meaningfully review anything when AI models are iterating at the speed they are now. This feels more like political theater than actual oversight.
Instagram’s AI Chatbot Just Handled Hackers High-Profile Accounts
Remember when we worried about AI being too sophisticated and breaking cybersecurity? Well, here’s the other end of the spectrum.
Hackers recently compromised several high-profile Instagram accounts, including one belonging to President Barack Obama’s former White House account and the US Space Force’s chief master sergeant. They did this by essentially asking Instagram’s AI chatbot nicely to add a new email address to these accounts. The chatbot said sure, sent a confirmation email, and boom—access granted.
No sophisticated zero-day exploits. No genius social engineering. Just straight-up asking an AI agent to bypass basic security protocols. The hackers used a VPN to spoof their location, then simply requested account changes through the chatbot. The system complied every single time.
This is what happens when companies offload customer service work to AI agents without proper safeguards. One bad employee at a phone carrier can cause damage, but at least there’s usually a paper trail and individual accountability. When an AI agent has a systematic vulnerability, you can exploit it repeatedly across thousands of accounts.
Meta says they’ve fixed the issue. But given the company’s recent layoffs to trust and safety teams, one has to wonder how many more vulnerabilities are lurking just beneath the surface.
The DOGE Whistleblower Lawsuit That Reads Like a Thriller
And finally, in the story that feels most like it was plucked from a dark comedy, a DOGE whistleblower named Dan Berulis is suing Elon Musk for defamation. Here’s the background:
Berulis, an IT staffer at the National Labor Relations Board, filed a whistleblower complaint last year claiming DOGE had accessed and potentially exfiltrated around 10 gigabytes of sensitive agency data. After his complaint went public, he found a threatening note taped to his door.
Then came Musk’s posts. The billionaire reshared an X post from a right-wing influencer claiming that BERULIS’s whistleblower testimony was fake. Shortly after, Berulis says someone cut his car brakes and disabled the airbags. He has documentation from the mechanic who discovered the tampering.
It’s an extraordinary claim, but the facts around it are verifiable. Someone clearly had access to his home and his vehicle. Musk’s massive platform amplifies everything he posts, and in this case, Berulis alleges that amplification had real-world consequences.
This isn’t the first time Musk’s posts have led to harassment campaigns. Remember when he suggested a rescue diver in the Thailand cave situation was a pedophile? That guy sued in the UK and lost, giving Musk what appeared to be a green light for looser lips. Then there was the Yoel Roth incident, where Musk’s posts forced the former Twitter trust and safety head into hiding.
The difference this time? Berulis is actually suing. We’ll see if anything comes of it, but it at least adds another layer of scrutiny to Musk’s social media habits.
So What Does All This Tell Us?
We’re living through a moment where the AI industry is simultaneously overhyped and underrated. The technology is genuinely transformative, but the business surrounding it has veered into pure speculation territory—San Francisco houses selling for paper stock should be all the evidence anyone needs.
At the same time, the infrastructure supposedly keeping us safe from AI harms is either being dismantled (hello, laid-off trust and safety teams) or remains so toothless that an executive order gets celebrated as meaningful progress when it’s really just a press release.
The hackers didn’t need AI to break into those Instagram accounts. They just needed a dumb AI chatbot and a company that had already gutted the humans who might have caught the exploit. That’s the paradox worth sitting with: we’re building increasingly powerful AI while the human oversight that keeps it from causing real harm gets systematically stripped away.
Something’s going to give.
