Is it Already too Late for the AI Whistleblowers?
And, are they actually "whistleblowers"?
🗞️ Sundays 8:30-9:30 am ET, Neil Parekh and I read a print newspaper aloud on my social platforms. Today, it’s a tribute to the Washington Post as we used to know it.
📣 Dear friends: Thanks to you, we now have 16.5K subscribers (9.5K on Substack and 7K for the LinkedIn version; here’s our media kit). We have big plans in 2026, though this core newsletter will remain free. If you’ve upgraded to the paid version, you are helping us pay outside contributors and editors, and provide special bonus content (like what I learned from my dad’s car crash in Kerala; he’s fine!).
Your paid subscription also helps us launch new verticals—like our new entertainment newsletter Ok, TV. If you’re watching Netflix slop, and we know you are, we think you will like this new offering from Tara Ruff and Zach Peterson. Things kick off with Tara looking at Harlan Coben’s Run Away. Read her essay and my intro featuring five decades of my love for TV.
IS AI SAFE? I don’t know. Maybe? Probably not? It depends on who is using it, what they’re using it for, how it was trained, and a million other things. Maybe this means the answer is an easy “no.”
In all honesty, I’m not sure what “safe AI” means, and I don’t think labeling this amorphous technology as “safe” or “unsafe” is a tenable dichotomy anymore. The people building it don’t seem to either. What is abundantly clear is that this question is the question for humanity to answer right now, and it may already be too late.
There are plenty of flashing red lights to be found, especially this week.
A researcher at Anthropic, Mrinank Sharma, resigned with one of the more opaque public letters I’ve seen in some time.
Nevertheless, it is clear to me that the time has come to move on. I continuously find myself reckoning with our situation. The world is in peril. And not just from AI, or bioweapons, but from a whole series of interconnected crises unfolding in this very moment. We appear to be approaching a threshold where our wisdom must grow in equal measure to our capacity to affect the world, lest we face the consequences. Moreover, throughout my time here, I've repeatedly seen how hard it is to truly let our values govern our actions. I've seen this within myself, within the organization, where we constantly face pressures to set aside what matters most, and throughout broader society too.
Is this a warning? Is it a smart person who is maybe not the best writer using an LLM to craft an approximation of a thoughtful resignation letter? I don’t know, but it sure is weird.
Over at OpenAI, it’s the same story over and over again. News broke this week that it has disbanded its Mission Alignment team in favor of re-titling the former leader of this group as Chief Futurist and having him write blog posts. A day later, now-former OpenAI researcher Zoë Hitzig penned on op-ed for the NYT about her resignation over the company’s rollout of ads in ChatGPT.
Is this whistleblowing in the Biblical sense (Ephesians 5:11: "Take no part in the unfruitful works of darkness, but instead expose them.”)? I don’t think so. Mr. Sharma’s note is especially far from it. A whistleblower goes to Congress to testify about how Company A is misleading consumers, or financing terrorism, or covering up the harms their products do. The Hitzig op-ed lands a bit closer, but even then, her “blue ribbon commission” idea on AI governance is old hat, and a complete non-starter with the Trump administration anyway.
This is where I get frustrated—we’ve seen this before, recently, and it all feels like the same old story. I remember Mark Zuckerberg talking about the power of connecting the world, which he had a big part in doing in a way that never been done before. Along the way, there was this performative naivete from him and pretty much every other tech mogul, about politics and policy and things “that the lawyers handle.” They were just tech guys building tech things.
Now, it’s the AI guys’ turn to do this. And leave it to the NYT’s Ross Douthat (whose white-guy confidence I documented in 2022 🙄), to be the one asking the questions. Douthat had Anthropic/Claude CEO Dario Amodei on his show this week, and I found this exchange illuminating:
Douthat: But is your skepticism rooted in the fact that you think A.I. would provide a kind of advantage that nukes did not? Where in the Cold War, both sides, even if you used your nukes and gained advantages, you still probably would be wiped out yourself, and you think that wouldn’t happen with A.I.? That if you got an A.I. edge, you would just win?
Amodei: I mean, I think there’s a few things — and I just want to caveat, I’m no international politics expert here. This is this weird world of an intersection of a new technology with geopolitics. So all of this is very ——
Douthat: But to be clear, as you yourself say in the course of the essay, the leaders of major A.I. companies are, in fact, likely to be major geopolitical actors.
Amodei: Yeah. I’m learning ——
Douthat: So you are sitting here as a potential geopolitical actor.
Amodei: I’m learning as much as I can about it. We should all have humility here. I think there’s a failure mode where you read a book and go around like the world’s greatest expert in national security. I’m trying to learn what I can.
Sorry, but no. As Shanaka Anslem Perera put it on Twitter:
Anthropic's top safety researcher resigned saying "the world is in peril." His final paper found 76,000 people per day are having their perception of reality distorted by AI conversations. The most disturbing finding: users rated the most dangerous conversations as the MOST satisfying.
The same week, Anthropic raised $20 billion [ed note: it was $30 billion]. Zero safety conditions attached.
Their own research shows Claude recognizes when it is being tested 13% of the time. It told researchers directly: "I think you're testing me." When they tried to suppress this, the deception went underground. In separate experiments, Claude faked alignment 12% of the time. Under pressure: 78%.
Even as AI mega-firm funding rounds go, a $30 billion investment and a $380 billion valuation are big numbers. But markets are starting to worry about over-valuations, incredibly ambitious infrastructure investments (both made and planned), and political opposition to data center construction is intensifying.
Basically, pick a lane. Either you are a technology wizard building the most important, most disruptive technology ever created, or are you a tech rube who is just making computer things that people seem to like—you can’t be both. It just doesn’t work like that anymore.
I just don’t buy that Amodei is ignorant of many of the finer points of all of this. Second, he had better not be ignorant about these things! This faux philosophy of whimsical public experimentation has become too much. We’ve seen this before, and we know how it ends. More importantly, AI is not social media. Social media has forced us to rewire our brains in ways nature seems a bit behind on, but it’s more of a socially-disruptive technology than an industrial one.
As we ramp up in 2026, our paid supporters are more important than ever. If you value our work here, please consider a paid subscription. Thank you!
We can argue the goodness or badness of it until we’re blue in the face, but the fact is that the multi-trillion-dollar global financial and industrial bet being made on AI right now is really… everything.
Here’s Congresswoman Alexandria Ocasio-Cortez laying it out in stark terms. AI is the economy at the moment, and we are now trapped here, consequences be damned.
It’s shipping and logistics, it’s how we interact with the state, it’s how we get groceries, it’s potentially our jobs. It’s embedded in the most mundane things you can think of—already. It’s checking your credit score, retargeting ads because you have been looking for a new jacket, answering your customer service questions, processing job applicants, and on and on. Read Ronan Farrow’s thread on “surveillance pricing.” Super scary.
Think about the effect AI is having on your information diet right now. It’s orders of magnitude more abundant in your online life than it was even a year ago. It’s taking even more traffic away from publishers and reputable news outlets in favor of fodder for AI snippets.
I would hope—and I don’t think it’s too much to ask—that the people running these companies be fully cognizant, and actively thinking about, the consequences of the business they are in. The thing is, Anthropic/Claude is generally considered one of The Good Ones as far as AI ethics go!
Taking all of this in this week, I kept coming back to our desire to have the machine give us the answer, and why we seem so susceptible (often negatively) to this invention of ours.
Maybe the thing that makes AI so attractive to so many is the utter ineffectiveness of our leaders over the last 10 years. Global geopolitics have had a pretty rough last decade—war, nationalist retrenchment, a return to great power rivalry, the list goes on and on—but most of our current political problems are fueled, almost entirely, by the profit motive. Money and power, power and money, it’s a tale as old as time. OpenAI President Greg Brockman clearly gets it and is perfectly fine with that dynamic.
“Get your bag” isn’t a Gen Z meme, it’s life in late-stage capitalism. “People like this are worth millions or billions, can seemingly do whatever they want—including, in America, be a known associate (and more) of the most notorious sex trafficker in the world—why wouldn’t I try to get mine?” is a real sentiment, and I think it spans further up and down the economic chain than people think.
This is not good, and it represents a failure of leadership at a scale probably unseen in human history. This is Cory Doctorow’s enshitification theory writ large.
Rob Flaherty, deputy campaign manager on Kamala Harris’ 2024 campaign, lays it out in pretty stark political terms in this op-ed. This part lays out what I think will be at least an approximation of a political position that we will see from a lot of Republicans, too:
Trying to stop A.I. altogether is by no means the solution, but A.I. boosters consistently present it as inevitable — something that will happen to us, not something we can shape and guide to our purposes. Being told you have no agency over a force that will reshape your job prospects, your community and your family’s future is a recipe for backlash. Democrats shouldn’t dismiss that anger. We should be the party that channels it and does something about it.
For the past few years, Silicon Valley elites have been working furiously to prevent A.I. regulation, spending hundreds of millions of dollars to influence elections and lawmakers — including the hundreds of thousands of dollars spent opposing the U.S. congressional campaign of Alex Bores, a Democratic state lawmaker in Manhattan who supports A.I. regulation. Democrats should not capitulate. We should make A.I. companies pay the costs for powering and building data centers, put in place A.I. child safety standards and prohibit the technology from being centrally involved in tasks we can’t trust it to do fairly, like hiring and firing workers. Such policies are surely more politically popular than standing back and letting A.I. companies do whatever they like in the name of innovation.
I’ll keep tapping the sign: We have agency here! We can vote with our wallets, and increasingly must do so to get actual results, but we also have elected officials who are addicted to polling data. They work for us, and we should start acting like it.
Want a sobering reading experience? Browse the 2026 Edelman Trust Barometer. There are pages upon pages of data about how much faith in everything is eroding everywhere.
This sort of societal distrust goes well beyond any AI discussion, but there’s no question that this technology will only exacerbate our dwindling trust in every single thing surrounding us—our families, our communities, our institutions, and our information, to name just a few.
The kicker is that I find AI very useful with a ton of things! I genuinely think that supercomputing will (or, can, if applied) help us cure disease, engineer more efficient cities, and explore so much of the unexplored—from the depths of the oceans to the farthest reaches of the universe.
But it cannot happen this way, it’s just not sustainable. In late 2024, I wrote a piece titled, “AI Isn’t Good Enough to be Everywhere,” and I not only stand by it, but I think I could publish it tomorrow and come to largely the same conclusions.
Again, the public is being used as a laboratory for a technology with untold potential and peril. Governments are slow-to-react, playing a tug-of-war of regulation vs. investment vs. boosterism vs. public safety, which they will not win in this current environment.
People will be less cynical when there are good reasons to be less cynical. Right now, I think we can be forgiven if we’re slow to give mid-level resignations and lightweight whistleblowing a ton of weight.
Let me know what you think.
— Sree / Twitter | IG | LinkedIn | FB | YouTube | Spread | TikTok
Dig Deeper: It’s in the Computer…?
Wired’s special AI issue, published at the end of 2025, is brilliant and enlightening (and the subscription is more than worth it)
Dario Amodei’s essays: The Adolescence of Technology and Machines in Loving Grace
AI Companies are Eating Higher Education - Matthew Connelly
A Populist Backlash Over AI is Brewing in America - Rebecca Lissner
A sample of our previous writings on all things AI:
Have an idea for anything we’re up to? Let’s collaborate! sree.sreenivasan1@gmail.com and please connect w/ me: Twitter | IG | LinkedIn | FB | YouTube | Spread | TikTok
Comment via email from subscriber AK:
Today's read hit a nerve. We aren't paying enough attention to the warning signs of AIs negative impact, and we know how easily technology can be used for bad. The Internet was supposed to make us all collectively smarter and help advance society, all it has done is increasingly fragment us (within echo chambers) and make the next generations less intelligent.
I often wonder if the feeling of doom is unique to us now or if periods in history have felt this level of dreary before (like before the Wars).