So… the AI world had another one of those weekends — you know, the kind where one scroll on X makes you wonder if robots are running politics now.
Let’s catch you up.
First up: last Thursday, OpenAI dropped what might be their most self-aware post ever. The company said, “ChatGPT shouldn’t have political bias in any direction.”
To prove it, they ran a massive “stress test” to see how objective their new GPT-5 models really are. Basically, they threw 100 political topics at ChatGPT — from immigration to abortion — phrased from both liberal and conservative angles to see if it leaned one way or the other.
The results? OpenAI says GPT-5 is officially the teacher’s pet of neutrality. Bias showed up “infrequently and at low severity.”
But here’s the spicy part — the test found that “strongly charged liberal prompts” made ChatGPT lose its chill a bit more often than conservative ones. So… the model’s not biased — unless, of course, you ask it spicy questions.
And they didn’t stop there. OpenAI revealed that the new GPT-5 Instant and GPT-5 Thinking models performed 30% better at staying objective than older versions like GPT-4o and GPT-o3. Now that’s impressive — though, considering the election-year chaos online, that’s like bragging your friend is “30% less argumentative” at dinner. Progress, right?
But while OpenAI’s trying to convince everyone it’s chill and neutral, another story blew up that’s anything but chill.
Lawyer Nathan Calvin, who works with AI policy group Encode, claimed that OpenAI actually sent a sheriff’s deputy to his door with a subpoena.
He says OpenAI wanted his private messages with legislators, students, and even former OpenAI staff. According to Calvin, the company used its lawsuit with Elon Musk as a pretext to intimidate critics pushing for AI regulation.
And OpenAI’s response? Chief Strategy Officer Aaron Kwon told The Verge the subpoenas were just about understanding why Encode joined Musk’s legal challenge — and that “deputies also work as part-time process servers.”
Which… is a very normal sentence if you’re definitely not panicking.
Then Tyler Johnston from The Midas Project — another AI watchdog group — said his team got similar subpoenas too. In theirs, OpenAI allegedly asked for “a list of every journalist, congressional office, partner organization, former employee, and member of the public” the organization has spoken to about OpenAI’s restructuring.
Even OpenAI’s head of mission alignment, Joshua Achiam, jumped on X to call out the move.
And honestly, when the guy literally in charge of “alignment” starts subtweeting his own company for being misaligned…you know things are getting complicated.
Meanwhile, as all this chaos unfolded, the rest of Silicon Valley was busy swapping name tags.
Andrew Tulloch, co-founder of Thinking Machines Lab — yep, the one Mira Murati left OpenAI to start — just joined Meta.
And Rishi Sunak — the former UK Prime Minister — snagged advisory gigs with Microsoft and Anthropic. Because apparently, retirement plans now come with a GPU cluster.
So yeah — between political bias tests, real-life subpoenas, and power moves that sound like Marvel post-credits scenes, the AI world had quite the weekend.
If any of these poked your curiosity, feel free to dig deeper.
