Anthropic’s IPO Could Put AI Safety to the Test

Knowledge at Wharton

Ask about Knowledge at Wharton

Episode Description

ABOUT THE EPISODE Anthropic’s potential IPO could be a major milestone for the AI industry, but can the company balance its commitment to safety with the pressure to compete and deliver profits?

Wharton Management Professor David Hsu discusses Anthropic’s IPO, the uncertainty surrounding AI business models, enterprise demand and the growing competition in frontier AI.

The conversation explores Anthropic’s unusual governance structure, its public benefit commitments, and whether AI safety can remain a priority when a company is accountable to public investors.

ABOUT THE PODCAST This Week in Business features interviews with Wharton faculty about the latest news, fascinating trends, and issues impacting both consumers and the business world. Episodes are recorded at the Wharton School and published twice per week.

----- The Wharton School of the University of Pennsylvania is committed to sharing its intellectual capital through Knowledge at Wharton, the school’s online business analysis journal.

Launched in 1999, Knowledge at Wharton offers free access to:

- Articles, podcasts, and videos highlighting Wharton faculty research and analysis of current business trends - Interviews with book authors from Wharton School Press - In-depth series of curated content like Wharton Executive Education’s Nano Tools for Leaders - A searchable database of more than 10,000 articles covering all aspects of business - A weekly newsletter that delivers Knowledge at Wharton insights directly to your inbox

Visit the Knowledge at Wharton homepage: https://whr.tn/3qUpSBE Subscribe to the Knowledge at Wharton YouTube channel: https://whr.tn/3DEAgAU

Read the full transcriptHide the full transcript
  1. Artificial intelligence company Anthropic is getting ready for what could be the largest
  2. initial public offering of stock ever, but it comes at a time where there are concerns around AI.
  3. There are comments recently by CEO Dario Amadei recognizing those concerns moving forward.
  4. So what does this S1 filing tell us about the future of open AI? Pleasure to be joined by
  5. Wharton Management, excuse me, pleasure to be joined by Wharton Management Professor David
  6. Hsu to discuss this. David, great to have you back with us. Good to see you, Dan. All right,
  7. tell me if I'm wrong about this, but part of this S1 filing is, as I said, the recognition that
  8. there are issues still out there and that need to be addressed. How unique is that to an S1 filing
  9. in the first place? In terms of the disclosures on the risks, you mean? Correct, yes. Yes, I think
  10. it's not unusual at all for a company in its S1 to really disclose everything under the sun. Now,
  11. what's important to realize is that it's not telling us anything about the probabilities
  12. of those risks, but as you'll note, there's plenty of risks that are beyond just anthropic and can be
  13. interpreted in terms of the evolution of the AI industry going forward as well. Right, and some
  14. of those we don't really truly have answers for right now, since this is still kind of a developing
  15. technology, whether it be those concerns about hackers, about agents, or whether it is the
  16. regulatory side, which we still don't have full answers for as well. That's absolutely right.
  17. This is an industry that we can certainly think about as clearly in the era of ferment,
  18. and as such there's a lot of question marks around the sustained demand for frontier models,
  19. for the development of the technology, and whether there's going to be slowdowns in the scaling properties as we feed it more data and more information,
  20. whether there's going to be a slowdown in performance. And probably most in my mind
  21. is the business model in terms of what customers are going to pay for, whether that will be
  22. sustained. And part of that answer also depends on the evolving competitive landscape. What are
  23. the alternatives? What are people going to be using it for? What are organizations going to
  24. be using the technology for? All those things are big question marks for not just anthropic,
  25. but for really this industry more comprehensively. So does that put in a little bit more of an unknown when you're thinking about those who
  26. might be wanting to buy on the IPO moving forward? Or is there enough of the belief
  27. of the industry about how it will be able to produce good and help workplace and so many
  28. other avenues that there is just this belief that taking part in this IPO is going to be a
  29. good thing even with the potential concerns? Yes, I think that this is what's I think a
  30. little bit unusual about this offering is this is the first time that the public retail
  31. institutional investors will be able to invest in kind of the frontier model layer. Now,
  32. I think that that's going to be exciting. But in my mind, the most exciting part is the part
  33. that happens after the IPO. That is much more information on a quarterly basis in terms of
  34. what is happening to enterprise sales, what is happening to those sales and customer demand
  35. with respect to the cloud subscription to enterprise customers, API usage, and moreover,
  36. how that compares to the alternative set out there. And so I think that there's a lot of real
  37. aspects to this that are just unknowable. That is, I'm sure that we'll get into it. But the
  38. element of the public benefits aspect and the safety aspects and how that plays into
  39. this worldwide race for the frontier technology and more importantly, kind of what's happening
  40. to these commitments that these companies have on the compute and infrastructure side
  41. and the time horizon of managing that in comparison to what's happening on a quarterly
  42. basis to the revenues that they're able to realize in the face of that evolving landscape.
  43. So, it was interesting you brought that up. I wanted to ask you about that next anyway.
  44. This quote-unquote long-term benefit trust that they've seemingly put into the documentation
  45. that they want to make sure that financial interests do not outweigh the public safety,
  46. which is a unique element to have into this, I would imagine. Absolutely. So, as you know, the founders retained 50.1% of the voting power. Then there's this
  47. public benefit aspect that will have a major role in selecting the board. And so this is
  48. clearly unusual in terms of the governance of something that's so frontier. But of course,
  49. it goes to the thesis of the company and why it spun off of OpenAI in the first place,
  50. which is putting into the constitution of training their models these elements of safety.
  51. Now, we've seen that all companies are not immune to these risks of going off the rails, etc.
  52. in the face of recursive self-improvement. And so I think it's going to be really interesting to see
  53. how this plays out between this kind of guardrail mission, public trust, and then this competition for leadership and being pioneering in these models,
  54. particularly when we have not just a domestic race on this, but also an international race in
  55. which we don't have a kind of governance framework on this. And so there's a lot of
  56. unknowns here with respect to how outsiders outside of the founders or the public benefits
  57. board will really interpret this mission of public safety vis-a-vis financial interests.
  58. And that does remain open to be kind of interpreted and seeing how this apparatus
  59. really behaves in as kind of scenarios unfold and how circumstances unfold in this landscape.
  60. Dario Amedi has been one of the people that has brought up the potential need to have a bit of a
  61. slowdown in the development of AI as we move forward. That's become obviously a very important
  62. topic. Does that have an impact on the IPO process by him having this belief and maybe that that is
  63. going to be a component of the future growth of open AI? I think it absolutely will, but in terms
  64. of predicting in what direction it's going to play out is not so easy. So let me explain that
  65. opinion in a little bit. So on the one hand, I think that it might be fair to say that together
  66. with the anthropic focus on coding, cloud, the enterprise segment, that's certainly an important
  67. reason for their ability to go public first ahead of open AI. But on the other side, there's a lot of
  68. narrative, both from the side of people, the people who might work at these companies,
  69. but also from the public, that anthropic has prioritized safety. And I think that that
  70. narrative, and that's played out. We saw what happened with the Pentagon, et cetera,
  71. with anthropic. But I think that what is unknown is, and it's hard for us to imagine without
  72. concrete examples, like what would anthropic do materially to really slow things down when you
  73. have these dual pressures? On the one hand, the need for revenue, especially salient when you're
  74. a public company. But on the other hand, how much would Amadei and anthropic tolerate in terms of
  75. the slowdown in terms of if there really is this trade-off between a slowdown in frontier model
  76. performance, of revenue growth, et cetera, versus what would be important in terms of taking time,
  77. safety. We've seen already that companies have decided, both open AI and anthropic, have decided
  78. to slow down ahead of the IPO on releasing certain models, et cetera, until we have a very firm
  79. governance structure in place. It's not exactly clear whose discretion it is in making that
  80. judgment. And so I think it is really difficult to say how this setup really will affect the IPO
  81. price. I tend to think about it more as more of a reaction once we see concrete test cases after the
  82. IPO in terms of market reaction to actions that are being taken by anthropic as it becomes a public
  83. company to these more tangible events that will happen. David, greatly appreciate your time and your insight on this. Thanks very much.
  84. Thank you so much. Take care. You got it. David Hsu, Wharton Management Professor, joining us here on the show.

Brief Summary

David Hsu says Anthropic’s IPO will test whether its AI safety commitments can hold up against investor pressure, competition, and the need for revenue.

Key points

  • The business model remains uncertain. Hsu points to open questions about sustained demand, model performance, and what customers will pay for. Those uncertainties affect not just Anthropic, but the broader AI industry.
  • The IPO’s real test comes afterward. Going public could give investors access to frontier AI, but quarterly disclosures will reveal more about enterprise demand, sales, and competition. Hsu says the bigger story is what happens after the IPO.
  • Can safety survive the pressure to grow?. Anthropic’s public-benefit governance is unusual, but its safety mission will face pressure from revenue goals and a global race to build frontier models. Hsu says concrete decisions will show how the balance works in pr

Tension

31/ 100Low tension

Related episodes