Robinhood CEO Says New AI-Driven Agent Apps Are Safe

Author: Bloomberg Television · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Robinhood CEO Says New AI-Driven Agent Apps Are Safe · Channel: Bloomberg Television · 07:03 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, AI safety

Summary

Robinhood CEO Vlad Tenev describes new agent features intended to connect AI models and external data sources with trading in Robinhood accounts. The company says agent apps may generate revenue through subscriptions, trading activity, and paid AI model usage. Users can place money in a separate agentic account, require approval for each trade, or enable autonomous agent loops. Asked about errors and liability, Tenev highlights engineering safeguards, trade approvals, and past use of agentic trading, while the interviewer notes that current disclosures place losses on users.

Key points

Tenev says Robinhood is building infrastructure for trading assets around the clock and integrating AI agents into its services.
Agent apps can connect trading strategies to external data sources, including Unusual Whales, Token Terminal, and satellite imagery.
Robinhood identifies subscriptions, trading activity, and AI model serving as potential monetization paths.
The company says its marketplace may offer discounted access to some app subscriptions.
An agentic account is a separate pool of money, which users must fund by moving assets into it.
Trade approvals are enabled by default; users can turn them off and set up agent loops to run strategies periodically.
The interviewer notes that Robinhood’s disclosure places liability for losses on users, and asks about agents making unintended trades.
Tenev says Robinhood uses defined tool calls and a harness layer to limit what agents can do, and is working on reliability and interpretability.
Tenev says more than 150,000 customers have used Robinhood’s earlier agentic trading product.

Chapters

Start	Chapter	Summary
00:00	Robinhood’s AI trading vision	Tenev describes Robinhood’s plans to connect AI agents with trading and external data sources.
01:39	Agent app monetization	Tenev outlines revenue possibilities from subscriptions, trading activity, and AI model usage.
03:18	Agentic accounts and trade controls	He explains the separate agentic account, default trade approvals, and autonomous agent loops.
04:51	Liability, safeguards, and reliability	Tenev discusses the risk of unintended trades and describes technical guardrails and prior use of agentic trading.

Quotes

we also have the ability to run agent loops. — 01:06

you can think of it as like a financial sandbox — 03:51

By default, trade approvals are on — 04:21

only allowed tool calls are are made possible. — 05:57

we’ve had over 150,000 customers using it — 06:29

Concepts and entities

Robinhood, Vlad Tenev, OpenAI, GPT-6 Luna, Robinhood agents, agent apps, agentic account, agent loops, Harness, hallucination

Full transcript (original language)

00:00 Joining us now live from that summit is the chairman and CEO of Robinhood, Vlad Tenev. Vlad, great to have you back on the program. You said last night that you’re basically putting the power of hedge funds into every trader’s pocket. Explain that. >> That’s right. I mean, we are in unprecedented times. You you hear that said a lot, but super intelligence is changing how we create and how we do everything. And Robinhood we’re focused on the financial aspect of this. And what what can happen

00:33 if you plug if you safely plug in the super intelligence into your Robinhood account. We take all the capabilities that Robinhood offers, which we announced 16 new ones last night, but think of it as the ability to trade any asset 24/7. And we’re really building the infrastructure for that. And with Robinhood agents, we can now plug frontier intelligence into all these assets and allow your intelligence to to

01:06 help you trade these things. We also introduced agent apps, which allow you to plug in unique data sources. So, things like unusual whales, token terminal, um satellite We have satellite imagery where you can actually tap into real-time satellite views of parking lots if you want to import that into your trading strategy. And and we also have the ability to run agent loops. So, these are autonomous runs where the agent can actually perform actions for you in your Robinhood account while

01:39 you’re sleeping. Uh so, we’re we’re very excited about what this unlocks. And it’s really just about making it clear to our active traders, who are the foundation of our business, that you’re at a disadvantage trading anywhere besides Robinhood. >> Well, I am curious specifically on the agent on the agent apps. I mean, how are you making money off that? Is that from subscriptions or or just a share of the agent app fees? I mean, I mean, what’s the what’s the direct uh financial benefit to you? >> Yeah, so there’s like three different potential monetization paths that you

02:12 can think of here. One is just subscriptions. So, these app developers uh are charging subscriptions and they’re actually offering deals where, you know, you get you get free usage for the first month. And, you know, in some cases through Robinhood agent apps, our marketplace, uh you can save money. Ra- It’s not just easier to integrate with Robinhood in your trading, but you can actually save money compared to getting the subscription directly or elsewhere. So, there’s a subscription and we take uh a piece of that for operating the

02:46 marketplace. There’s also just the trading activity, which we monetize normally on our platform uh and and we get that. And then, for now, uh we’re we’re offering free usage uh through OpenAI of GPT-6 Luna till the end of the year. Uh but, there’s also monetization on the AI model serving layer if you’re using, you know, one of the one of the premium models that cost money. >> Talk a little bit more about what the users are actually going to see when

03:18 they use this um trading agent. Do they just see the completed order or do they get a little bit of insight into the rationale behind, you know, every single trade? What is the model looking at? Um what kind of risk limits is this agent looking at? >> Yeah, and, you know, it’s if you uh we gave a pretty cool demo of all this. So, if the viewers want to get in deeper, we I encourage you to check that out. But, basically, what this is is it’s a separate account. So, for those of you that have Robinhood, we allow you to

03:51 separate your assets into multiple accounts. Of course, we have your main Robinhood account, but we also have retirement, we have joint accounts, we now offer trusts, and we’ve introduced an agentic account, which is like a separate money bucket away from your retirement, away from your main brokerage account, and you can choose to move money into an agentic account, and you can think of it as like a financial sandbox, you know, to make sure that your agent doesn’t go and just trade all of your money. You have to make

04:21 affirmative moves to actually move money and assets into the agentic account. By default, trade approvals are on, so you’ll you’ll approve every trade that the agent wants to make. But if you want, you can turn that off, and then you also have agent loops, where you can essentially create uh loops where the agent periodically does does its analysis, runs the strategy, and decides whether to trade, and that allows you to operate your agent even when you’re not

04:51 looking at your Robinhood app, like when when you’re sleeping um because now, you know, equity markets are going to be 24/7 starting with Robinhood first, so you’ll always be able to capture opportunities. >> Well, I do I want to just ask you though about the liability, and I did look at the disclosure, and it’s very clear that the dis- uh that the the liability risk, at least for right now, uh for losses still rests with the the user rather than Robinhood itself, but have you uh talked with your tech team about this idea of whether the agent uh and and the technology behind it maybe hallucinates

05:24 uh and does the wrong ticker or or basically does something that the user did not intend it to do, does that loss or potential loss still fall on them, or would Robinhood actually bear uh some of that uh some of that responsibility? >> Yeah, I mean, uh I guess without going into like the legal nature of this question, uh what I’ll tell you is we spent a lot of effort and time making sure we mitigate the risk of hallucination and not not just hallucinations in the core models which

05:57 have been improving, but also making sure that it calls Robinhood services through well-defined tool calls. And and there’s actually a lot of engineering on what’s called the harness layer which is the layer that goes around the model and connects it to our system to make sure that only allowed tool calls are are made possible. We spend a lot of effort, you know, making that really really robust and making sure that you know, the the reliability is is up

06:29 to our standard. And you know, again, there’s trade approval so you can monitor what the what the agent is doing. There there’s also we’re investing more in predictable determinism so that the the outputs are interpretable. And we should also say that this isn’t our first agentic product. We rolled out agentic trading a few months ago and we’ve had over 150,000 customers using it and actually building things. And this is a an extension of that. Of course, it’s going to be much

07:03 bigger now that it’s inside the Robinhood app, but I think we’re building increasing confidence that these tools are not just powerful and open new capabilities, but but they’re also they’re also safe and we believe that through our guardrails we can actually introduce these products to users in a way that is safe and and also useful.