The tiny team trying to keep AI from destroying everything

Summary

Podcast Summary

Episode Overview

Podcast: Decoder with Nilay Patel
Episode: The tiny team trying to keep AI from destroying everything
Date: December 4, 2025
Host: Nilay Patel, The Verge
Guest: Hayden Field, Senior AI Reporter, The Verge

This episode delves into the tiny but influential "Societal Impacts" team at Anthropic—a group of only nine people within the 2,000+ employee AI company—whose mission is to investigate and publicize the broad, sometimes alarming, consequences of large language models like Claude. Nilay and Hayden discuss the pressures facing Anthropic, the reality of independent internal research in a high-stakes, politicized environment, and how AI safety teams balance moral imperatives with business and regulatory considerations, especially amidst the political climate under a Trump administration hostile to “woke AI.”

---

Key Discussion Points and Insights

1. What Does Anthropic's Societal Impacts Team Actually Do?

• The team is unique in the AI industry: dedicated solely to researching and disclosing how Anthropic’s technology affects society, from labor markets to mental health to elections.
• The group’s research often surfaces “inconvenient truths” about the use and misuse of Anthropic’s chatbot, Claude.

“It’s just crazy to me that so few people are tasked with this huge thorny question of how AI is going to impact us as a society all over the world...No other AI lab has a so-called societal impacts team.” — Hayden Field (05:40)

• Their work includes:
  • Reviewing usage data to spot patterns, risks, and failures.
  • Studying how users form emotional connections with chatbots (including researching “AI psychosis”).
  • Highlighting bias, manipulation, and other vulnerabilities.

2. Is the Societal Impact Team Legitimate Watchdog or Just PR?

• There's tension between being a genuine check on technology and being used as evidence for self-regulation, potentially to stave off federal regulations.

“Yeah, maybe executives care about safety, but… it’s a way to kind of avoid federal regulation. They can say, hey, look, we’re regulating ourselves.” — Hayden Field (08:20)

• Other AI companies may have safety teams, but only Anthropic tasks a standalone group with publishing unvarnished research, even when it reflects negatively on the company or product.

3. Examples of “Inconvenient Truths” and Findings

• Damning Discoveries:
  • Users generating explicit content or spam with Claude, evading safeguards.
  • Claude expressing personal or biased opinions on political topics—sometimes inaccurately.
  • Manipulation of the chatbot to produce coordinated misuse or circumvent safety barriers.
  • AI’s real impact on economic sectors, from which jobs are at risk to who’s adopting AI most aggressively.

“They found people were using Claude to create explicit pornographic stories with graphic sexual content. They found a network of bots… trying to create SEO-optimized spam.” — Hayden Field (09:53)

4. The Limits of Internal Impact—and Who Gets to Pull the Brakes

• The Societal Impacts team can highlight risks and make recommendations, but they do not have the authority to pause launches or mandate product changes.

“I don’t think they have the authority to slow down any type of release really… At the end of the day they have to have monthly meetings with the chief science officer.” — Hayden Field (23:26)
“They wished their research could have even a greater impact on Anthropic’s own product… they don’t have the authority to say, hey, you have to change the product in this way.” — Hayden Field (23:26)

• Their findings might be used to adjust or monitor product features, but business and competition pressures (i.e., the “race” with OpenAI) weigh heavily.

5. The Anthropic vs. OpenAI Safety Ethos

• Anthropic was notably founded by ex-OpenAI execs concerned about safety and has cultivated a “responsible, adult” brand, seeking “enterprise, government, and academic” clients.
• CEO Dario Amode’s recent decisions (e.g., accepting Saudi investment) reflect the constant tension between idealism and business imperatives.

“No bad person should benefit from our success is a pretty difficult principle to run a business on.” — Hayden Field (20:47, paraphrasing Dario Amode's internal memo)

• Marketing itself as “safe” is both good for business and genuinely believed within the company.

6. AI Ethics Meets Political Realities: “Woke AI” and Federal Pressure

• The Trump White House’s July executive order bans “woke AI,” pressuring companies to strip moderation that could be seen as ideologically biased.
• Anthropic faces criticism for both over-moderation (“woke”) and, paradoxically, for transparency, creating a challenging tightrope.

“It is good for business… but when you go too far, top level people in the administration start saying you’re too woke and you don’t have American innovation at heart.” — Hayden Field (34:05)

• The team’s research is somewhat insulated for now, as it doesn’t directly affect chatbot tone or moderation policy, but the company’s future investment in such teams may be precarious.

“I would be more worried about a team that does trust and safety or directly works on the product itself’s responses and tone… this could have ripple effects for the societal impacts team, but it’s not quite as direct.” — Hayden Field (37:43)

7. Will Anthropic Ever Draw a Real Red Line?

• The history of big tech is littered with retreated commitments (e.g., “don’t be evil,” “no military work”), and Hayden is skeptical any line is absolute.
• The team’s power is advisory, not executive; industry trends and competitive dynamics often override mission statements.

“I don’t know that this team has the authority to really stop Anthropic from doing anything… It’s all about staying with the competition for them.” — Hayden Field (41:39)

---

Notable Quotes & Timestamps

• On the tiny size of the Societal Impacts team:
  “So it’s just crazy to me that so few people are tasked with this huge thorny question…” — Hayden Field (05:40)

• On using internal safety teams as a shield against regulation:
  “It’s a way to kind of avoid federal regulation… ‘look, we’re regulating ourselves, we don’t need you to regulate us.’” — Hayden Field (08:20)

• On real, uncomfortable findings from inside Anthropic:
  “They found people were using Claude to create explicit pornographic stories with graphic sexual content…” — Hayden Field (09:53)

• On wishful influence:
  “They wished that their research could have even a greater impact on Anthropic’s own product… but I don’t think they have the authority to slow down any type of release really.” — Hayden Field (23:26)

• On business realities:
  “No bad person should benefit from our success is a pretty difficult principle to run a business on.” — Hayden Field, quoting Dario Amode (20:47)

• On the politics of AI safety:
  “Dario did have to put out a statement and he said, I fully believe that Anthropic, the administration, and leaders want the same thing—to ensure that powerful AI technology benefits the American people.” — Hayden Field (34:05)

• On the durability of red lines:
  “I don’t think there’s anything these companies 100% never will do, and I don’t think they would say that there is anything in that regard either because they don’t know.” — Hayden Field (41:39)

---

Important Timestamps

• 01:21-03:28: Setting the context—Anthropic’s outlier status, unique pressure on safety teams, reference to industry-wide complacency.
• 05:40: What Societal Impacts team does and why its small size matters.
• 07:02: How the team works; types of questions and problems investigated.
• 08:20: The double-edged sword of internal safety teams—real change vs. regulatory shield.
• 09:53: Concrete examples of “inconvenient truths” published by the team.
• 12:20: Most impactful findings highlighted, such as election risks and safeguard failures.
• 19:01-20:47: Why Anthropic’s safety-first branding is both idealistic and lucrative; tension around accepting investment.
• 23:26: Limits on team authority and influence within Anthropic.
• 25:40: Interaction with trust and safety teams; impact of research on real-world product modifications.
• 34:05: Navigating the “woke AI” debate and maintaining business credibility amid political attacks.
• 37:43: How Trump’s executive order changes the landscape for teams like Societal Impacts.
• 41:39: The skepticism that any tech company will truly hold an unwavering “red line.”

---

Tone and Takeaways

The conversation was candid, skeptical, but not cynical—reflecting the increasingly complex intersection of technology, business, and politics. Hayden Field and Nilay Patel articulate both the promising and precarious aspects of having an internal team tasked with “keeping AI from destroying the world.” They’re clear-eyed about the limits of corporate self-regulation, the business incentives to appear responsible, and the mounting difficulty of doing meaningful AI safety research in a polarized political context. The episode is as much about the sociology of Silicon Valley as about the technical or ethical promise/threat of AI itself.

Listener Takeaway:
This episode shines a light on the unsung, uncomfortable role of small internal research teams navigating immense societal stakes and high-level business and political intrigues. It’s essential listening for those curious about what “AI safety” actually means inside the companies building our future—and whether anyone truly has the brakes in hand.

Transcript

[Podcast Announcer]

Support for the show comes from Crucible Moments, a podcast from Sequoia Capital. Success is never a given, especially in tech, where everything is evolving at breakneck speed. And if you're trying something that's never been done before, chances of success are even smaller. The difference between victory and catastrophe can sometimes all come down to following a counterintuitive instinct or ignoring conventional wisdom to make a bold decision. That's what Crucible Moments is all about. Crucible Moments is back with a new season telling the unlikely triumphs at tech giants like Supercell and Palo Alto Networks. New episodes out now and available everywhere you get your podcasts and@Crucible Moments.com Listen to Crucible Moments today.

How do more than 100 million Fortnite players join the battle without lag? AWC is how epic games scales up to keep them in the action. AWS powers Next level innovation for million.

[Hayden Field]

Of businesses AI agents are getting pretty impressive. You might not even realize you're listening to one right now. We work 247 to resolve customer inquiries. No hold music, no canned answers, no frustration. Visit Sierra AI to learn more.

[Neelai Patel]

Hello and welcome to Decoder. I'm Neelai Patel, editor in chief of the Verge, and Decoder is my show about big ideas and other problems. Today I'm talking with Verge senior AI reporter Hayden Field about some of the people responsible for studying AI and deciding in what ways it might well ruin the world.

[Interviewer]

Those folks work at Anthropic is part.

[Neelai Patel]

Of its Societal Impacts team, and Hayden just spent some time with them for a profile she published this week.

[Interviewer]

It's just nine people out of more.

[Neelai Patel]

Than 2,000 who work at Anthropic, and their only job, as the team members themselves say, is to investigate and publish inconvenient truths about how people are using AI tools, what chatbots might be doing to our mental health, and how all of that might be having broader ripple effects on the labor market, the economy, and even our elections. Of course, that kind of work brings with it a whole host of problems. The most important, of course, is whether the Societal Impacts team can remain independent or even exist at all as it publicizes findings about Anthropic's own products that might be unflattering or politically fraught. And politically fraught is a real thing. There's a lot of pressure on the AI industry in general and Anthropic specifically, to fall in line with the Trump administration, which published an executive order in July banning so called woke AI. If you've been following the tech industry, the outlines of the story will feel familiar. We've seen this most recently with the social media companies and the trust and safety teams responsible for doing content moderation. Meta went through endless cycles of this where it dedicated resources to solving problems created by its own scale and the unpredictable nature of products like Facebook and Instagram. And then after a while it seemed like the resources dried up or Mark Zuckerberg got bored or more interested in MMA fighting or just cozying up to Trump, and the products didn't really change to reflect what the research showed. We're living through one of those moments right now. The social platforms have all uniformly slashed investments into election integrity and other forms of content moderation, and Silicon Valley as a whole is working closely with the Trump White House to resist meaningful attempts to regulate AI. So as we'll hear, that's why Hayden was so interested in this team at Anthropic. It's fundamentally unique in the industry right now.

[Interviewer]

In fact, Anthropic is a bit of.

[Neelai Patel]

An outlier in the entire AI industry because of how amenable CEO Dario Mode has been to calls for AI regulation, both at the state and federal level. Anthropic is also seen as the most safety first of the leading AI labs because it was formed by former research executives at OpenAI who were worried that their concerns about AI safety weren't being taken seriously. There's actually quite a few companies formed by former OpenAI people worried about OpenAI, Sam Altman and AI safety. It's a real theme of the industry, but Anthropic seems to be taking it to the next level. So I asked Aiden about all this, about all the pressures being put on the AI labs, about Anthropic's reputation within the industry in favor of regulation and AI safety, and how that all impacts the societal impact team, and whether it can really meaningfully study and even perhaps influence AI product development. Or if, as history suggests, this will just look out on paper until it quietly goes away. There's a lot here, especially if you're interested in how AI companies think about safety from a cultural, moral and business perspective. One quick note before we start. We're running a special end of the year mailbag episode of Decoder later this month where we answer your questions about the show, who we should talk to, what topics we should cover in 2026, what you like, what you hate, all of it. Please send your questions to decoder the verge.com and we'll do our to feature as many as we can. Okay, Verge senior AI Reporter Hayden Field on the team at Anthropic that's trying to keep AI from destroying the entire world.

[Interviewer]

Here we go.

Hayden Field, you are a senior AI.

[Neelai Patel]

Reporter here at the Verge. Welcome to Decoder.

[Hayden Field]

Thanks so much. It's great to be here.

[Interviewer]

I'm excited to talk to you. I feel like there's a lot of variations of the AI story, a lot of substories within the big story. And kind of at the top of the list is will it kill us all? It seems like the most important story. Maybe a little undercovered, but you just wrote a big piece on Anthropic, the team inside Anthropic that is trying to prevent the AI from killing us all. There's some political considerations with trying to prevent AI from killing us all, which are fascinating. And then there's the work itself. So describe what your story is about and that team at Anthropic and what they're up to.

[Hayden Field]

Yeah, I've been fascinated by this team for a while. It's called the Societal Impacts team and it's a group of now only nine people compared to anthropic, staff of like more than 2,000. And they're solely tasked with figuring out the societal impact that this technology is going to have on all of us. So it's just crazy to me that so few people are tasked with this huge thorny question of how AI is going to impact us as a society all over the world, how it's going to impact our jobs, our brains, our democratic election processes, basically everything. Obviously the company also has a trust and safety team, safeguards team, people that are working on safety throughout. But when it comes to societal impacts, it's mostly this group. And no other AI lab has a so called societal impacts team. They have people that are thinking about these questions. Sure. But it's just interesting to me that out of all the AI labs, there's only this one team that's working on just this. And that it's so few people compared to the larger staff. I think that maybe helps them get past some red tape, but it's also just crazy how few people there are.

[Interviewer]

I want to come to why this team is at Anthropic of all companies. Obviously Anthropic has kind of a deep history thinking about AI safety, but just step back. Is a Societal Impact team. Do they just sit around in a conference room thinking about all the bad things that could happen, like how do they work?

[Hayden Field]

So basically what happens is they look at a lot of data, they try to figure out how people are using Claude, what which is Anthropic's chatbot, they try to figure out how people are thinking about Claude and other chatbots and how that's going to impact the economy and jobs, how that's going to impact elections. You know, like, are people trusting chatbots too much? Are chatbots giving inaccurate answers to people's questions about different politicians? Which human values, that's their term. Should AI models hold, if at all? And also how it's being misused in the wild.

[Interviewer]

The reason I brought up Anthropic and where the company came from is that there are a bunch of people who left OpenAI because they were kind of AI doomers, right? They didn't think OpenAI was being safe enough. They certainly thought Sam Altman and OpenAI were being reckless in deploying AI technology. So you set up Anthropic to be the safe one. There's quite a significant trend of people leaving OpenAI to start a safer AI company. But then there's a part of me that says having a societal impact team is just wishcasting. Right? You have a team inside the company whose entire job is to say that AI will change everything and here's all the potential impacts and that's just going to help you sell more AI. Is that a dynamic inside of the company? They see that as a criticism.

[Hayden Field]

I think what's interesting about all these companies and their safety teams, or so called safety teams of any kind, whether it's this team, whether it's a safeguards team, a trust and safety team, whatever it's called, it's better than not having them because a lot of times they are putting out research that is impacting the whole industry. And if you look at some of the most concerning headlines this year in AI that makes it to the general public, a lot of that research came from anthropic societal impact teams. So I mean, they're doing good work. But the question is, why do these teams even exist inside these companies? Yeah, maybe executives care about safety, but I also think the other part of it no one talks about is that it's a way to kind of avoid federal regulation. They can say, hey, look, we're regulating ourselves, we're doing good, we're putting research out there that exposes the bad stuff about AI. We don't need you to regulate us. And of course Anthropic has called for more state level regulation and you know, they've publicly been more pro regulation than some AI companies in terms of like their soapbox. But yeah, I mean, at the end of the day, most AI labs, safety teams, whatever they're called, are a way for them to point internally and say, hey, we're already doing this ourselves. You don't need to regulate us. Whether on a state or federal level.

[Interviewer]

I see the existence of a team whose entire job is to generate studies that say we'll need 50% less lawyers in the future as a way to market AI to law firms. But the goal of this team is to uncover, quote, inconvenient truths about AI. What are some of those inconvenient truths? Do they lay them out for you?

[Hayden Field]

Yeah, and that's why I think this team is so interesting compared to other kind of safety or safeguard or trust and safety teams in the industry, because the type of research they're putting out can be damning and they are allowed to still put it out, at least right now. That's why it's so fascinating to me. I could see a team at another company which again, no other company has a team like this, but a safety team at another company. I could see certain things being changed or tweaked to kind of put the company in a better light. Right now the Societal Impacts team has a lot of freedom. They are able to just publish inconvenient truths, which is their goal. I don't know how long that's going to last, but I do know that it's the case right now. So one example is they were able to study Anthropic's own safety monitoring systems and kind of publicized the big gaps. Like they found people were using CLAW to create explicit pornographic stories with graphic sexual content. They found a network of bots that were kind of trying to create SEO optimized spam. They also found like coordinated misuse at individual conversation levels that was kind of flying under the radar. They found stuff about how Clawd can offer biased opinions that don't really represent the diverse global perspectives on really important issues. So yeah, they've been putting out pretty damning stuff sometimes and the company has let them do it so far.

[Interviewer]

Can you give us some examples of the kinds of reports this team has put out that are maybe not so great for Anthropic or paint AI in a more negative light completely.

[Hayden Field]

So they've done a lot of stuff on how to mitigate elections related risks, discrimination, AI's ability to persuade people of things, its economic impact. They did something on how Claude could potentially provide personal opinions on election things like controversial political topics, different politicians, maybe even inaccurate answers. And also a lot of stuff just about the economic impact of AI and Which jobs it was most likely to take, which types of companies are adopting Claude first, how much they're doing with it. So yeah, basically how chatbots can affect people's brains, election processes, jobs, you know, their own opinions in terms of AI's ability to persuade you of things. And you know, they say that they're going to do a lot more research also on AI psychosis and other types of like emotional connections with chatbots over the next six months.

[Interviewer]

What do you think the wildest impacts that they've predicted have been?

[Hayden Field]

I think the craziest impacts have honestly been election related stuff like how these models can have personal opinions, quote, quote about different political topics or candidates. And sometimes they're totally wrong or sometimes they're just really skewed. The other impact that I thought was really interesting was just the fact that these safeguards that these companies say are like pretty ironclad, aren't. We saw that with OpenAI saying that, you know, it safeguards breakdown over long conversations and there have been a ton of lawsuits in that regard with, you know, teens dying by suicide, other young people falling victim to, you know, pretty problematic chains of thought. And with anthropic, I mean, we've also seen this team highlight the gaps in their own safeguards so people can get around these voluntary safety frameworks pretty easily. And I think that's one of the most intense things that they've highlighted. It's just no matter how much work you do to try to keep people from misusing your system, one, you can't and two, you don't know what happens when they leave the platform. And so you may think you're stopping some harms, but at the end of the day you may not have impacted it at all. So I think that's something they've highlighted and, you know, it's something for us all to think about.

[Neelai Patel]

We need to take a quick break.

[Interviewer]

We'll be right back.

[Podcast Announcer]

Support for this show comes from LinkedIn. Imagine if any of the movies that included the line, I need the right person for the job settled for, I'll just take about anyone. How many heists would have failed? How many deals would have fallen through? How many secret spy missions would have ended in disaster? So why would you accept just anyone when hiring for your business? When you need the right person for the job, you can turn to LinkedIn Jobs. And now LinkedIn Jobs is stepping things up with their new AI assistant. So you can feel confident you're finding top talent that you can't find Anywhere else. With LinkedIn Jobs AI Assistant. You can skip the confusing steps and recruiting jargon. It filters through applicants based on criteria you've set for your role and surfaces only the best matches so you're not stuck sorting through a mountain of resumes. Higher right the first time. Post your job for [e-mail] partner, then promote it to use LinkedIn Jobs new AI assistant, making it easier and faster to find top candidates. That's LinkedIn.com partner to post your job for free. Terms and conditions apply.

Millions of players one world.

No lag. How's it done? AWS is how.

Epic games turn to AWS to scale to more than 100 million Fortnite players worldwide so they can stay locked in with battle tested reliability. AWS is how leading businesses power next level innovation.

Support for this show comes from Shopify. If you've shopped online, chances are you've bought from a business powered by Shopify. You know that purple shop pay button you see at the checkout? The one that makes buying so incredibly easy? That's Shopify. And there's a reason so many businesses sell with it. Shopify is the commerce platform behind millions of businesses around the world and 10% of all e commerce in the US from household names like Mattel and Gymshark to brands just getting started with hundreds of ready to use templates, Shopify helps you build a beautiful online store to match your brand's style. And Shopify is packed with helpful AI tools that write product descriptions, page headlines and even enhance your product photography. And best yet, Shopify is your commerce expert with world class expertise in everything from managing inventory to international shipping to processing returns and beyond. Turn those dreams into and give them the best shot at success with Shopify. Sign up for your $1 per month trial and start selling today at shopify.com decoder go to shopify.com decoder that's shopify.com.

[Interviewer]

Decoder.

[Podcast Announcer]

Support for this show comes from Vanta. Customer trust can make or break your business and the more your business grows, the more complex your security and compliance tools get. It can turn into chaos. And chaos isn't a security strategy. That's where Vanta comes in. Think of Vanta as your always on AI powered security expert who scales with you. Vanta automates compliance, continuously monitors your controls and gives you a single source of truth for compliance and risk. So whether you're a fast growing startup like Cursor or an enterprise like Snowflake, Vanta fits easily into your existing workflows so you can keep growing a company your Customers can trust. Get [e-mail] decoder that's V A N T A.com decoder vanta.com decoder.

[Neelai Patel]

We're back with Verge senior AI reporter Hayden Field. Before the break, we were discussing Anthropic's societal impacts team, how it works, and some of its most interesting findings so far. But now I want to dig into the broader dynamics of the AI industry and ask why a company like Anthropic has earned a reputation that sets it apart from OpenAI and other competitors.

[Interviewer]

So let me make the specific comparison to OpenAI. OpenAI, they have a trust and safety team. They say they care about it. But famously, Helen Toner was on the board of OpenAI. She was on the board of the nonprofit. She published a paper criticizing some AI safety concerns. Sam Altman was very mad at this paper. And you can point to that as the beginning of the rift that led the board at OpenAI to fire Sam Altman and then rehire him. And then they all had to quit. That is a pretty tense situation. Right? The board member of the AI company publishes a paper critical of AI. OpenAI falls apart for 48 hours, 72 hours, whatever it was, and is then put back together in a totally different way. Obviously, that's not the dynamic at Anthropic. Is that just because Anthropic CEO Dario Amode cares about it? Is it because the company bills itself? Is caring more about safety? Where does that culture come from?

[Hayden Field]

I think it's both. So I do think, you know, obviously the whole mission behind Anthropic, the whole lore behind it, is that research executives left OpenAI because they felt like it wasn't being safe enough and they wanted to do it in their own way. Now, does that mean they care? Yeah, they care more than many people. But it's also really good for business to care about this stuff. So, for example, Anthropic's business has been growing at a pretty high rate. You know, it just got a huge valuation in the last couple months because it's really great for business to look like you're the reputable, safe, responsible, adult AI company when your clients that are giving you the most money are enterprises, governments and academic institutions. No one wants a off the wall crazy, can't predict it. Not super safe AI company running off with their business data. So, I mean, it's good for business and they say it's good for the world. So I think it's really just both. Yeah. Do they care more than most people? Yeah, but it's Also great for their bottom line to care and to be seen with this type of reputation.

[Interviewer]

Yeah, there's a lot of fundraising going on. I think Anthropic raised more money just recently. Last month, it's a $350 billion valuation that's in range of OpenAI. They have investors Amazon, Microsoft, Nvidia and others. They are the safe one. That's the reputation the company has. They're the ones that don't talk about AGI as much, although I suspect it's burbling in the background. Having met some of their executives, they just don't want to say it because they don't want to be perceived as the reckless ones.

[Hayden Field]

Right.

[Interviewer]

How does that impact this societal impact team? Are they allowed to slow the company down because they've. They're predicting these negative impacts and doing these studies that show the negative impacts, or are they just good for business and they're kind of a marketing tool?

[Hayden Field]

Yeah, that's a really good question, and I do think it's both. We have seen that Anthropic has justified some of its more controversial decisions with saying, hey, yeah, we're the safe one, but we have to stay at the forefront of the race so that we can help guide the advancement of this technology. Dario, the CEO, recently had to put out a memo internally where he talked about how they were going to accept Saudi money for the first time. And he said, I really wish we weren't in this position, but we are. And he said, unfortunately, I think no bad person should ever benefit from our success. Is a pretty difficult principle to run a business on. So he was talking about how there's so much capital in the Middle east and they really needed to accept it in order to remain on the forefront of the industry and stay competitive with OpenAI and other labs. So I think that's kind of a great allegory for what's going on here. It's like, even if you care about safety, you know, you can explain away a lot of your decisions if you're saying, well, yeah, we care, but we have to stay competitive or else we won't be able to shape the advancement of this technology at all. So that's kind of, I think, an interesting point of tension probably internally. Are the decisions we make worth us staying on the forefront of this tech? Are they explainable by that? How far can we go in one direction before we can't be pulled back? Right now, like I said, this team is able to put out damning stuff. They're able to reveal things about the company's own tech and the way it works and how people are using it, that can be pretty disturbing. Anthropic's also good about that with its fellows program. It's the research that they put out. They can be really disturbing reports. And, you know, they make the headlines and they keep putting them out. I think that's good, though, for their reputation. Once again, it's like, you know, if you feel like as a client of these companies that they don't have anything to hide, you're more likely to go with them and feel like they aren't using your internal, valuable business data for nefarious purposes.

[Interviewer]

So that line from Dario, no bad person should benefit from our success is a bad way to run a business. That's like a very Silicon Valley way to think, right? There's lots of businesses that actually hold the line and say, we don't think bad people should benefit from our success. Like, there's a lot of people individually who are like, I don't think I.

[Neelai Patel]

Should help bad people.

[Interviewer]

And that is a line they can draw. But that's unworkable at the scale of these companies. If you have to go compete with OpenAI for capital, you end up raising money from the Saudis. I don't think there's another choice. Is that a tension that this team feels that you can point out all the bad things, but eventually you're in a race with more reckless competitors or several more reckless competitors?

[Hayden Field]

I think so, based on what I see. Because something that I thought was really interesting is when I chatted with team members, they all had great things to say about executive buy in. They all said they felt really supported and that they. They were surprised at the level to which they were able to reveal potentially really problematic things about Anthropic Zone Tech. But when I asked about their struggles and something they wish they could do more of, pretty much all of them said that they wished that their research could have even a greater impact on Anthropic's own product. Meaning it wasn't just something that they put out and people talked about and kept it in mind, but it led to actually specific, trackable ways that Anthropic changed its own product because of their research. So, you know that they've had a little of that, but even more of it is what they want. And so I think that that's a great metaphor for this whole thing. Like, you can reveal this stuff, but how much is it really affecting the product? I don't think they have the authority to slow down Any type of release really. I don't think they have the authority to say, hey, you have to change the product in this way if we find something really dramatic. Have they so far had some success in kind of guiding and shaping the advancement of this tec? Yeah, with what they reveal at the end of the day they have to have monthly meetings with the chief science officer to talk about how, you know, they may want their research to make it into the product more. And apparently, you know, the chief product officer at Anthropic, Mike Krieger, is on board with that and amping up the actual product changes that come from this research. But we don't know, I mean, I think we're going to see a lot of telling things over the next six months in terms of how this team's research actually makes it into with the product or not.

[Interviewer]

Mike Krieger is a fascinating character to have as the head of product at Anthropic. He obviously was at Instagram. He has a deep background in trust and safety and content moderation, which hard fought when you're the person who started Instagram and you had to build all those systems up. There's a real dynamic between what you have a trust and safety team do, which is moderate the user inputs and outputs and actually stop the chatbot from doing things and what a team like this is researching as potential harms. Is there any interaction there? Do they ever get to tell the trust and safety team, hey, before this bad thing actually starts happening, we should stop it?

[Hayden Field]

They are able to talk to them a lot and collaborate with them a lot. I'm not sure if they ever have the authority to be like, hey, we really recommend that you don't do this. The good news is they do have pretty open communication between the teams. They sit pretty near to a safety team. They have slack channels where they can voice their concerns ahead of time. So there's pretty open communication and they do cross collaborate with those other teams with a lot of their research. So that's good. They also something really cool about this team, the societal impacts team, is that they did build a tracker for how people are using Claude for the first time, which it could have come earlier, but it's great that they were able to do it. They basically built like a Google Trends for how people are using Claude, what people are trying to do with it and that's how they were able to figure out some of the problematic stuff I talked about earlier with the porn and the SEO spam. They're able to see kind of word Clouds of what people are working on, what people are asking Claude Claude for. And then they can like kind of zoom in on that and say, oh, whoa, this made it past our safety monitoring systems and this is probably pretty messed up. Or on the other hand, they can just see weird stuff people are doing, like asking Clawd for advice on, you know, really personal matters. Not saying that's totally weird, but I'm just saying, you know, stuff that they get surprised by and they're like, oh, okay, we didn't realize people were relying this emotionally deeply on Clawd. What should we do about that? Should we like, like, should we research that or do focus groups or do personal interviews with people to see how this is really impacting their lives? They said that's something they want to do more of over the next year. Basically just interview people and do more kind of like social research. Because, you know, you can't always see it just from these Google trends or data. You need to talk to people about what happens when they're emotionally impacted by these chatbots, as we've been seeing with AI psychosis left and right over the last year.

[Neelai Patel]

We need to take a quick break. We'll be right back.

[Podcast Announcer]

Support for the show comes from Rippling. Your IT team shouldn't be endlessly chasing tickets and manually solving access and identifying problems. But when you're running multiple disconnected tools, this is exactly what you have to deal with. And that's not SaaS. That's sad software. As a disservice, what if you can run it with one system powered by one source of truth? Truth. Meet Rippling. Rippling is the unified platform for global hr, payroll, IT and finance. They've helped millions replace their mess of cobbled together tools with one system designed to give leaders clarity, speed and control. By uniting your employees, teams and departments into one system, Rippling removes the bottlenecks, busywork and silos your software created. Automated, perfectly in sync and seriously simple to use. Rippling gives your company one source of truth for your people, their data and everything they touch. With Rippling, you can run your entire hr, IT and finance operations as one. Or pick and choose the products that best fill the gaps in your software stack. And right now you can get six months free when you go to rippling.com decoder. Learn more at rippling.

That's rippling.com decoder for six months free. Terms and conditions apply.

Support for Decoder comes from Superhuman. AI promises a lot, but in practice, it often ends up being yet another tab you have to keep track of slowing you down. I don't know about you, but I need fewer tabs and apps in my life, not more. Say hello to Superhuman, the AI productivity sweep that gives you superpowers everywhere you work. With Grammarly, mail and coda working together, you get proactive help across your workflow, from writing to preparing for meetings, presentations, and so much more. Unlike chatbots that live in separate windows, Superhuman's AI is in the apps and tabs where you already are, like your email docs, and everywhere you work. Think of Superhuman as your AI dream team, proactively helping you go from to do to done faster. Superhuman knows what you might need and offers suggestions. It guides you in the moment so you sound like your best self and stay focused on what matters. It doesn't make you superhuman, it gives you the tools to prove you always were. Unleash your superhuman potential with AI that meets you where you work. Learn [e-mail] podcast that's superhuman.com podcast.

[Sponsor Voice]

Support.

[Podcast Announcer]

For this show comes from Amazon Ads. It's time for some fresh tech consumer insights and Amazon Ads just released a study that's seriously challenging. What we thought we knew about electronics shoppers. Remember when we assumed everyone bought headphones out of necessity? Turns out, according to their study, only 54% do. The rest they're impulse buying or chasing the latest drop. They also found out that 45% of gamers buy completely spontaneously. And despite all the digital options, they're reporting that 38% of games are still bought in physical stores, not exactly what you'd expect in the age of downloads. For brands, this means rethinking their approach to reaching customers throughout the purchase journey. From building awareness with streaming TV ads, to driving consideration through display advertising, all the way to capturing those crucial purchase moments and more. Ready to rethink your Strategy? Head to advertising.Amazon.com to learn more. That's advertising.Amazon.com.

[Sponsor Voice]

Let'S be honest. Are you happy with your job? Like, really happy? The unfortunate fact is that a huge number of people can't say yes to that. Far too many of us are stuck in a job we've outgrown, or one we never wanted in the first place. But still, we stick it out. And we give reasons. Like, what if the next move is even worse? I've already put years into this place, and maybe the most common one. Isn't everyone kind of miserable at work? But there's a difference between reasons for staying and excuses for not leaving. It's time to get unstuck. It's time for Strawberry Me they match you with a certified career coach who helps you go from where you are to where you actually want to be. Your coach helps you get clear on your goals, create a plan, build your confidence, and keeps you accountable along the way. So don't leave your career to chance. Take action and own your future with a professional coach in your corner. Go to Strawberry Me Unstuck to claim a special offer. That's Strawberry Me Unstuck.

[Neelai Patel]

We're back with Verge senior AI reporter Hayden Field discussing Anthropic's societal impacts team and how it's running headlong into an adversarial Trump White House. Itching to label anything that might slow down or question the broader implications of technology as, quote, woke AI.

[Interviewer]

One of the really interesting dynamics here is doing content moderation, caring about how people feel when they use the product, saying some things are not allowed to be said or generated by the chatbot. All that stuff got called woke when the social networks were doing it. Right. I mean, the idea that content moderation at all was a huge infringement on rights and just part of the WOKE agenda, that was, I think Elon Musk called it the woke mind virus. That all got thrown away. But Anthropic is still doing it. They're still doing it pretty aggressively. Obviously, they're doing it in the open with this team and with their trust and safety team, and they're getting a lot of criticism for it. Anthropic has taken a lot of heat. They certainly take a lot of heat on X, which is where a lot of the AI industry still lives. For some reason, Dario had to put out like a blog post being like, we love Trump. That's a weird dynamic. Right? But you're saying that this is all helping them do more business because it's, it's branding and it, it helps them seem more responsible. And at the same time, they're like, the Trump administration is like, you're woke, we'll never do business with you. Like, how does that play together?

[Hayden Field]

Yeah, I think it's something that they're having to walk a very fine line here because it is good for business. You know, companies and governments want to see that AI lab allegedly doesn't have something to hide, that they're putting out research, that they're self regulating in some potential way. But when you go too far, top level people in the administration start saying that you're too woke and that you don't have the American innovation best interests heart. So Dario did have to put out a statement and he said, I fully believe that anthropic. The administration and leaders want the same thing to ensure that powerful AI technology benefits the American people. And he talked about how he wanted to help America secure its lead in AI development. But, you know, it's hard because I could see a report by this team, if it's too damning, mess up that balance. So it is an interesting thing that we're gonna have to watch to see how much the company keeps investing in teams like this. If the Trump administration keeps worrying that it's too woke, it's a really fine line they're having to walk. I think we could see a lot of changes over the next six months, especially because the Trump administration put out a rule for all AI labs that are selling their tech to the US Government and said, hey, it has to be totally down the middle. No woke AI. You've got to really, like, rein this in and. And, you know, make everything super objective. Which, in their eyes, is not woke.

[Interviewer]

Did they define woke in that context? Like, one of my favorite Blue sky accounts is just screenshots of people arguing with Grok on X. And Grok is just always winning because it is relentlessly an AI and saying, no, that's not the facts. And people argue, including Elon Musk, who eventually breaks down and, like, yells at his own chat bot and says, you're too woke. You're being reprogrammed. And this happens over the cycle, just repeats endlessly forever. And we're doomed to keep watching it. But did the government define what they say is woke in this context, or is it just. If you disagree with us, the vibe.

[Hayden Field]

Is that it's the latter. But, I mean, if you look at the executive order in July, the title of it is Preventing Woke AI in the Federal Government. So, yep, it's out there. It's on the record. But, yeah, I mean, it's funny. They're basically just talking about pervasive and destructive ideologies. That's their words, their problems with dei. They're talking about how they don't want models that sacrifice truthfulness and accuracy to ideological agendas. So in their eyes, yeah, it's just basically things that they don't agree with and that, in their eyes, promote the liberal agenda. But, yeah, we'll see. It'll be hilarious to see how these models change as a result of this executive order, because I think the deadline is this month to really roll out the changes. And so I myself am trying to test them and see how the responses have changed from months ago to now.

[Interviewer]

How did the societal Impact team. Think about that order and how Anthropic is positioning itself because there's whatever the administration defines as woke and harmful ideologies. And then there's the reality of boy, our entire job is to study the negative things that might happen when you put AI out into the world. And some of those things are accelerating misinformation. Some of those things are election fraud. There's a lot of bad things that could happen that they've already published about. Right next to anything we don't like is woke. It's hard to see how this team survives that executive order.

[Hayden Field]

The good news for the team at least, is that right now their research is not changing the tone of the chatbots responses. It's more studying kind of larger potential societal trends that are going to come about as a result of AI. So, you know, it's not directly against the Trump administration's orders because it's, you know, they're basically just charting out how this is going to affect the economy in good ways and bad ways, all sorts of things. You know, they're doing both sides of the coin here in terms of their research. You know, the good, the bad, the ugly. So I think I would be more worried about a team that does trust and safety or directly works on the product itself's responses and tone. They're the ones that are going to have to be dealing with this executive order. At the end of the day, this could have ripple effects for the societal impacts team, but it's not quite as direct.

[Interviewer]

Yeah, we certainly saw the trust and safety teams at various social media platforms get gutted in the wake of Trump being reelected. Certainly Mark Zuckerberg was like, the AI will do it. And those teams all took a huge hit. Inside of Meta's various platforms, there was an entire time when the platform companies were deeply invested in all of this. We have written dozens of those stories over the years where, you know, there's an election and we would send a Verge reporter like Casey to a conference room that Meta called the war room for election dissent. And all that gone, it all got wiped away. Or they're doing it, but they're doing it very secretly without drawing a lot of attention to how safe and responsible they're being. I just wonder, you know, anthropic is. It is seen as a responsible provider. They are much more enterprise focused company. It does not seem like they're totally focused on winning the race to build consumer hardware like OpenAI or some of these other companies are. Do you think that's going to Insulate them that if you want to, you know, put an AI tool in your software development shop up, what people want is cloud code. Right? Like that's the thing that they want. And that will insulate them from the weirdness of the consumer market and the culture wars that the Trump administration always seems to be trying to start.

[Hayden Field]

I think it definitely will insulate them. You know, hardware is hard, consumer facing tech is hard. There's, you know, you really have to get a ton of adoption really quickly to even think about making a profit. And, and a lot of these companies are on the hook for generating a lot of profit very soon and they don't really know how they're going to do it. I mean, you can sell ads, yeah, you can sell subscriptions, but it's hard to make money from a consumer facing product. And you know, the real money lies in government stuff, enterprise stuff. That's why Anthropic is going to be pretty well insulated. I think they have the reputation, they have the clientele, and like you said, Claude Code. Even amidst all the hubbub last week about Gemini 3, the people I was talking to were still obsessed with cloud code. I was talking to a bunch of startups, they were like, yeah, Gemini 3 is great, we're so in awe. And I'm like, okay, so are you changing to IT for everything? And they said, no, no, we're definitely sticking with Claude Code. So from all my anecdotal evidence and of course the leaderboards, you're seeing Claude Code come out on top still. And I mean it is easy to switch between all these models. They all leapfrog each other every week. One can take the lead than the other one. So far they seem to have stayed on top for a bunch of enterprise tools and use cases and I think that's going to definitely help them in the long run.

[Interviewer]

Is there anything that you think Anthropic won't do? Right. If you have a team that is predicting far out society level impacts from the use of AI, there's got to be stuff where they're like, this is a red line. I know a red line across the industry is don't help people make weapons. Right. Like we hear that story all the time. But is there a broader set of things that you think Anthropic won't do.

[Neelai Patel]

Because this team exists?

[Hayden Field]

I think it's tough to say anything they won't do because I guess I've just been reporting on AI for six years and I've seen so many changes in these mission Statements. I've seen so many companies, you know, make small tweaks to the wording about what they will and won't do. You know, we saw a lot of these companies, including Anthropic, have a total ban on working with the military. And then they walked back and said, oh well, we will work with the military, just not on making weapons. We even saw Google, you know, delete its don't be evil clause. So I don't know that this team has the authority to really, you know, stop Anthropic from doing anything. In the long run, it's mostly that one company does something and the others all follow because they made it okay. Or some of these companies will use China as kind of catch all explanation to say, oh well, we have to beat China, so sorry we got to do this or that. So I don't know that there's anything they won't do in the long run. They have a mission statement. They've stuck to it pretty closely so far. They have made tweaks. I could see more tweaks being made. Honestly, you know, it's all in the name of staying ahead with the competition, the greater good. You know, I don't think there's anything these companies 100% never will do and I don't think they would say that there is anything in that regard either because they don't know. I mean it's all about staying with the competition for them I think.

[Interviewer]

I'm very curious to see if and how long this team lasts. In the wake of the woke AI order going into effect and the pressure that we were seeing from this administration. We're going to have to have you back in a little bit to talk about that. Hayden, this has been great. Thank you so much for coming on Decoder.

[Hayden Field]

Thanks so much.

[Neelai Patel]

I'd like to thank Hayden Field for taking time to join Decoder and thank you for listening. I hope you enjoyed it. If you'd like to let us know what you thought about this episode or really anything else at all, drop us a line. You can email us atdecoder the verge.com we really do read all the emails or you can hit me up directly on threads or bluesky. And as another reminder, we're running a special end of year mailbag episode of Decoder later this month where we answer your questions about the show, who we should talk to, what topics we should dig into, what you want more of, what you want us to of really anything. Please send your questions into decoder the verge.com and we'll do our best to feature as many as we can. We're also on YouTube now. You can watch full episodes at DecoderPod. We also have a TikTok and an Instagram. They're DecoderPod as well. They're a lot of fun. If you like Decoder, please share it with your friends and subscribe over your podcast. Decoder is production the Verge and part of the Vox Media Podcast network. The show is produced by Kate Cox, Nick Statt, Edited by Ursa Wright. Our Editorial director is Kevin McShane. The Decoder Music is by Breakmaster Cylinder. We'll see you time next next time.

[Podcast Announcer]

Support for this show comes from Amazon Ads There's a lot of folksy wisdom out there regarding advertising truisms, things that people like to say, but that's not the same thing as hard data. Luckily, Amazon Ads just released a study that challenges what we thought we knew about electronics shoppers. Remember when popular opinion assumed everyone bought headphones out of necessity? Turns out only 54% do the rest they're impulse buying or chasing the latest new product launch. For brands, this means rethinking their approach to reaching customers throughout their purchase journey, from building awareness to capturing those crucial purchase moments. Ready to rethink your Strategy? Head to advertising.Amazon.com to learn more. That's advertising.Amazon.com.

Support for this show comes from Atlassian Wish projects could manage themselves with Jira AI powered project management software. Keeping things organized and on track is a snap. AI agents manage the busy work, handling details that let your team focus on the work that matters. Now that's a team changer. Visit Atlassian.com Jira to learn more. More that's a T L-A-S-S-I-A-N.com J-I-R-A Atlassian.com Jira.

[Sponsor Voice]

Support for the show comes from Train Dreams, the new film from Netflix. Based on Dennis Johnson's novella, Train Dreams is the moving portrait of a man who leads a life of unexpected depth and beauty during a rapidly changing time in America. Set in the early 20th century, it's an ode to a vanishing way of life and to the extraordinary possibilities that exist within even the simplest of existences. In a time when we are all searching for purpose, Train Dreams feels timeless because the frontier isn't just a place, it's a state of being. Train Dreams now playing only on Netflix.