Name: Claude for Financial Advisors: Your Cowork Security and Compliance Questions, Answered
Uploaded: 2026-10-20T17:34:45.646Z
Duration: 57 min 44 s
Description: Claude for Financial Advisors: Your Cowork Security and Compliance Questions, Answered

Transcript for "Claude for Financial Advisors: Your Cowork Security and Compliance Questions, Answered": Hello, everybody. Thank you for joining. We're very excited about today to to just give some intro and background context. About a month ago, we launched the Claude for Financial Advisors plugin, and we have had a ton of, demand and interest about... On the tool. There have been over 60,000 installs at individual organizations across the world, which has been fantastic. There have also been a ton of questions around Cowork security, info security, data retention, and all of the the topics. And we thought it would be most helpful to get everybody into the same room today and get you all the questions that you need to feel comfortable in turning on each of these tools. But I'll go to the next slide here. Just in terms of quick housekeeping, this will be recorded, so don't feel like you need to scramble to take notes. We'll get this recording out to everybody that's registered. Please submit your questions at any time to the q and a. We're holding time at the end to just go through anything that's posted there, and give us feedback as well. I think, it's... We obviously don't get to communicate directly on these webinars, so please just let us know how we can continue to improve this experience. But in terms of what we'll cover today, we'll we'll give just a brief overview on both what Cowork is, and how it differs from just your typical chat interface in Claude. I'll also give just a brief refresher and overview of the Claude for financial advisors plug in and what that is. We will go into a deep dive into the actual Cowork controls and security features, enterprise readiness, etcetera, and then we'll have Q&A at the end. Just to give a a quick intro, my name is Drew Parker. I head up the RIA go to market, function for Anthropic, and I'm joined by Dor Fledel. Hey, everyone. Dor Fledel, thanks for having me. I'm Dorud. Did cyber my entire life, and now really helping with the enterprise readiness for cybersecurity on the product side. So really helping our customers adopt our products in a secure way. Amazing. Yeah. So I think we... We'll start with a with a quick poll and, and we'd love to see where, this this crowd is. What percentage of, your advisers use Cowork or any other agenda tool? I'm not taking anything personally, but, but, really excited to hear. So for example, if they're just using chats today, or similar, that that doesn't count. So take a few more seconds. We already see some votes coming up. Yeah. Very interesting. So what we're seeing is that, about half, use, of of the people in the crowd think that in their, like, employee base, thanks for the people who voted, who made it more than half, are... Don't use a cork or any other agenda tool. And I'm... We're seeing a a few that that actually, like, have more than that. Drew, I'm curious from from what you're seeing with your, your customers and across the industry. How does that, like, align? Yeah. I I think that's about right. I think, there's a lot of uncertainty and unknown with regard to how can a firm that is highly regulated turn on Cowork and Agenca capabilities in a safe fashion. A lot of those questions have frankly stemmed from, product features that we need to to ship after initially, releasing Cowork. But the good news is is that we have everything that we need today to... In order for the most regulated firms in the world to turn this on. Okay. So I think, to start, it would be helpful just to give, like, a quick refresher and for those that aren't familiar, overview on the Cloudstra Financial Advisors plugin, which lives in Cowork. And then we'll we'll, go on into just, like, a general Cowork overview. But CloudFront Financial Advisors is a plug in. What does that mean? It's just a package or bundle of connectors to the wealth tech tools. So think, portfolio management systems, financial planning tools, alternatives, CRMs, etcetera, so that you can basically plug Claude into the tools that you're using today. So Claude becomes a centralized intelligence layer applied across the entire stack. On top of each of those connectors, we prebuilt and shipped skills or really workflows that you can run with on Claude. So there's a pre meeting skill, a post meeting skill, and investment proposal skill, a portfolio rebalance skill. For context, the skills are incredibly generic in nature intentionally. We we did that so that the tool and the skills are as usable by the widest array of firms possible. And our hope and intent is that each and every one of you in your firms take these skills and customize them the last sort of 10 or 15% of the way there so that they're ready for production workflows, because every firm operates a little differently. Every firm is their slightly own different interpretation of the marketing rules, and you all have your own disclosures that you need added and brand guidelines, etcetera. And that's all great context that can be layered into Claude. To. give just an over... Oh, go ahead, Dor. Yeah. Yeah. I I can share at least on on our side. On the product side, I have, like, monthly metrics report that I, that we took, like, a skill, then slightly adjusted the 10% which you're saying, and that really saved me, like, hours a week of reformatting title, headers, and the font text and stuff like that. Yeah. 100%. So now just, like, general overview on Cowork. The main difference between Cowork and Chat is the chat experience is prompt and response. You're not gonna have Claude take an entire goal for you or an end state and reason its way past multiple different decision points and tool calls that it needs to make, etcetera, to give you finished work as an output. It is, just basically a one to one experience. Whereas Cowork, you can simply just give Claude the goal that you're looking to have it complete, and Claude will determine the most optimal or best way to get to that goal and look across the different MCP connections that you have stood up to make tool calls and pull reports or draft items. And then you can also have Claude spin up agents to review all of the... This work as well. And what you're left with is finished work as an output. This becomes incredibly important too as you start to look towards the future to where this is all going. So when we sort of first came into, like, the world of AI, for public use, the models were not particularly intelligent, and they weren't able to run for long durations of time, on a given task without major issues. What we're seeing is with every subsequent model release, intelligence obviously exists on this exponential curve that we've all come to understand, but so does the, harness or agent's ability to perform a single task for a given duration of time that also increases. And so what you're gonna be left with in a future world is incredibly intelligent and powerful models that can go out into the world and do tasks that could last for a quarter at a time. And so I think what we're seeing with some of our maybe more unregulated customers that have had access to cohort for a longer period of time because they haven't had the same, compliance and infosec constraints that everybody on this call has today is they have been able to take advantage of the intelligence that they have access to right now, give it, give the agent or Cowork a prompt that requires a lot of work, and Cowork can run on that all the way to completion, for very long durations of time. And as every subsequent model is able to run for longer and is objectively smarter, the art of the possible just continues to exist on that exponential. And I think we talk a lot internally about how this all sort of started with prompt engineering in chat where you needed to give the context in every single prompt to Anthropic to, like, refine the actual output that you're giving. Then that changed to, like, skill engineering and creating repeatable skills that you can invoke consistently or repeatably, to have the same output given. Then it kinda became harness engineering. Where does it all end? Goal engineering. Just give Claude the ultimate goal that you're looking to do and let it run for the duration of time that it needs to, and get the output. And I think the advantage to firms that really figure that out is is you can have Frontier Intelligence look into your business, look into your products, look into whatever you're doing, and make... And discover findings and find optimizations for processes and other things that you just, like, haven't even thought of yourself, because it has all this ability, connectivity to your stack, context on your firm, and etcetera. So I think that's why Cowork becomes really important. I think it's also, I'll switch this slide here. This all sort of, like, culminates in, like, okay. That's great. But, like, is this safe to turn on? The answer is yes. I think when we initially launched Cowork, it was in a beta state. And, frankly, it didn't have all the features from an enterprise readiness, info security, and, like, audit logs, etcetera, that regulated firms needed to turn it on without scrutiny. We were very transparent about that and blasted it out with the release saying like, hey. If you're regulated, this probably isn't good to go for you right now. However, we have shipped a ton of changes over the past couple of months that we're gonna highlight here today. Cowork is now in a state where it can be used by any firm entity company in the world regardless of regulation. And we'll we'll go into detail into all of that, and please give us your questions. But, I think what we need to do today is just get everybody to a yes on, like, feeling comfortable to turn it on. And this is also incredibly important timing as since, when we initially released Cowork, it wasn't ready for regulated firms. We made the decision to separate the chat and Cowork interfaces. Going forward, starting in next year, we are going to be unifying and combining the chat and Cowork interface into one. And so our goal today is get everybody comfortable with Cowork so that we can get that enabled for firms so that when the unification happens at the beginning of next year, it's a seamless transition, and nobody feels like they're being left behind. But with that, I will turn it over to Dor. Yeah. Incredible. Thanks for the context, Drew. So, I'll focus on on security controls and really piggyback on whatever Drew just said. Like... And I want to give you an illustration of what we had and what we have now. As you can see in the beginning of the the year, we, the enterprise readiness wasn't there, and we're very transparent about it as as Drew said. And since then, a a lot of work, prompt, tokens, and agents, really worked on on making this a reality. And what you can see here is how much progress, we've made. I would say that, typically, when I talk to customers, they're used to release cycles that take, like, three to six months from the second, that there is a problem even to a point where someone even talks to them about the requirements, let alone start coding that. And that to itself takes a year. Here, development cycles are very, very fast. That's part of the AI revolution and how we use AI, so on and so forth. So what it puts on you, sadly, but but fortunately, is, that reevaluation needs to to happen in a much faster pace than than what you're used to. If in... With other SaaS products, evaluating it in May, dimming into a certain result doesn't really apply now five months later. And we do feel very comfortable about what we provide. So to give you slightly more sense, Drew Drew it up, and, I would love to make sure that the audience here is fully aware because we, think that this is an important evolution of what we have. We have two modes of co work. One that we call local execution and the other that we call cloud execution. The local execution is what the 99% of the people in the enterprise world are, aware and used to. In the consumer world and pro and max, this is already, gone, and they use the bottom, thing. So we've done to allow long running tasks and to allow users to not depend on desktop apps and similar. We actually have done a... An architecture change where we move part of the execution to the cloud, as you can see on the on the bottom sketch, the bottom two sketches. The main thing that I would love you to take is that we're now living in a dual mode. And as Dor said, coming end of this year, beginning of next year, more and more would, transition to the cloud hosted execution. From an end user person, we'll go into the benefits in a second, but just know that from an end user perspective, the move is seeming... Is seamless. Like users, your users would not really feel it. And we just want to be very, upfront and very direct about the upcoming switch, especially as we now see more and more customers transition to it because we actually think that there are, like, material upsides to it. So let's focus for a second. In general, and we'll share share the slides and and the recording later, but we've even created, like, a diagram for you on how things are said, what can you put where different controls, and we'll go over it in detail in a second. No need to memorize it. But when we're thinking about the remote execution, what we see from a security perspective is that we see a contained blast radius because we run each session in their own VM instead of on your devices. We see that as a IT and security teams, you can enforce centrally, policies and not do them per device or do them via MDM and similar or proxies as firewall similar, less put... To protect on the endpoint. And it also allows us to use state of the art network lockdowns and additional things that we can do on our cloud while you cannot do on your cloud. I would also say it really stems from a better user experience. Like, what really users told us is that they can now finally shut their laptops off. And, and that was material. And so if they allow a long running task, scheduled tasks. It allows if a user just change... Forgot their laptop somewhere or change the laptop, it allows them to retain their history. And also potentially relevant for some of you, it allows VDI, and virtualization because in that case, you... We keep the security and safety while keeping that. So in general, there are a lot of upsides. The end user transition is seamless, and I want to reemphasize that. We just wanted you to be fully aware as we're now diving into the topics. So what are the existing controls? And I think this... If there is one one slide that I would love you you to take from this one is is this one. All of those are are ready for you, already today. So we have a single sign on and RBAC where you can control who gets what in a very grounded way on, on a product level. You can, for example, start a pilot on core... For just a subset of your team. You can start the remote execution pilot for for a subset of your team. You can allow, certain types of connectors for a subset of your team. And that's really granular and really evolved over the last months as you might expect. Network controls really allowing, the sandbox, to really constrain off, like, blocked by default. You can allow it in and also adding IP allow lists in case you want to connect it to your on prem and network in a very specific way and so on. On data controls, and I think that for, for what I'm hearing from customers, data is is material for for them as AI is definitely like a new, a new concept in the market. Everything is encrypted at rest. If some of you follow a certain, regulation, you can also ask your account representatives for a CMEK, the customer managed encryption keys that's also available for you today. We can have custom data retention. The average customer we have has, the median has a ninety days of data retention, which means that if a chat was not active for more than ninety days, it is deleted. In that sense, you limit your exposure, with all data that is retained, but in the same time, balance the productivity of your team. Because sometimes people like to search. Think about, like, your email search that, you sometimes use. Auditability is a big thing we improved. We released compliance API, and we'll go, into that for Cowork and any future, like, the CloudAppSecured Cowork as well. And really reemphasizing something that, comes up a lot in conversations, but fairly easy. The no training commitment. We do not train for teams and enterprise. It's in your... You can see it on our pricing page. It's in the customer agreement. We totally understand the issue there, and, and that's why we we fully commit to it. So you can, I know that sometimes there are, like, extra fear about those things, but that's that's part of our, our terms and legal terms? Before we can dive deeper into each one of those, Drew, anything. to add? Yeah. I think that this is amazing, Dor. Thank you so much. I think a question that I often get from, RIAs and other wealth management firms is in regard to data retention versus data isolation and the difference. I think there's a lot of uncertainty and maybe even confusion around the difference and what firms need to be looking for, whether they need zero data retention on their enterprise license in order for it to be compliant or not. I think any context that we can provide there would be super helpful. And on top of that, and just for context, everybody on this call is dealing with client PII and very sensitive information that needs to be locked down and secured. So, yeah, I would love to to double click there. For sure. So I I I want to jump right into it and and, and let's, let's make sure that we we address the the data separation and isolation, as well. But we've built a multilayer approach to to data security. It starts with both compliance API, which is a retroactive control that allows you to see every prompt, response, and material data. We'll share more in a second. And Inference Hooks, which is going GA today, which is, generally available today, which allows you to actually also block in line. And it's an revolutionary protocol, and concept that we developed here, and it is now reaching maturity and gets very positive traction. On top of it, the encryption keys can be yours, can be ours, but encrypted by default on every tenant regardless of of which tenant you're on. And prompt injection and a lot of safety there and logical isolation we're doing between different, customers. And on top of that, we all know that you have, some security vendors, maybe more, maybe less. We have the widest catalog of security integration in the industry, more than 100 security, integrations where you can already consume the security value with your trusted vendors and just plug them in in a simple and easy way. So on top of that, to go back to to Drew's question, each one of those multilayer, defense really shifts the the concept between, like, let... What we call in the industry isolated tenants for your firm versus an isolated, organization. We've built our software in a fairly modern stack, unlike other software that you might use to. And therefore, the problems of noisy neighbor, data, data, flows between, between different organizations. They all don't exist because things are are, fairly modern here. And and that's why as a customer, I would like you to move away from being about, like, having your own tenant versus having what this outcome tries to avoid... Tries to achieve, which is exactly what the controls and the architecture that we built, that each core session has its own VM, its own session, its own logical isolation, really, really helps. If we move forward, I want to address the third row there, which was about prompt injection. You can see public data that we that we released about prompt injection. I would say that many security and IT practitioners I'm discussing with, have it as like a top of mind, item, which is totally understandable. We've taken this feedback in our research, in our labs, how we, evolve the model and really made it materially harder and nothing is impossible. So I would, say just materially harder, and you can see here how much, to prompt inject our, models and our models are trained for it. You can also see how they're compared to the industry in open source models. Whoever of you who... Who's thinking about open source models for your clients' data, and we're taking that to heart. So I wouldn't say that this is now, impossible, but I would say that we've put a lot of effort into it. And I think there is a lot of, like, former concepts and knowledge that were correct a year ago. But as we put SAFE in the top of our minds, we really put the prompt injection as a as a metric that we're evolving the models towards. So really, alleviating a lot of concerns by enterprise customers once they see this data. If we move forward, specifically to touch upon Cowork, a lot of customers are asking us about local data, local folders, what can it access, whatnot. So I want to be slightly prescriptive. We even show the the MDM flags that you can use. But first and foremost, they allowed folders and allowed the allowed data. By default, the users, like, core work is is not accessible to anything. That's the default. What your end users need to do, they need to actually practically allow each folder and each file. So core work doesn't mechanically do whatever it wants unless you allow it to do. With that in mind, admins can also set certain constraints around access to folders. In that sense, you can, for example, what I've seen another, big firm do is just allow one specific folder that all Windows laptops come mounted with, and that specific folder is accessible to Cowork. With that in mind, that really limits and puts into a certain framework and swimming lanes what you allow your end users to do and kind of like how much you make it proof, for for any human error. You can also limit whether it's read mode, read and write, so on and so forth. From data movement perspective, there is a... There are two options. For Cowork that runs on the local device, we purposely beat Cowork to make the data movement efficient and lean. What does that mean? Let's say you give it the the bible. It doesn't really send the bible to Anthropic, if you ask it to summarize the bible. It actually goes through very select chapter and so on and so forth and transfers to the inference only a limited amount of context. The main reason is because we want to make it token efficient, network efficient. We want to make it fast, and you don't really need, to transfer the entire bible or 10,000, files of your employee case... Of your client cases. So in that sense, only data that is processed is being sent from your user's laptop to our cloud, and the exposure is much more limited than than, the the entire phone. On the local device and execution, that's an admin setting. It really allows, for example, what we've seen a lot of customers do is, like, automate browser use, for applications that are not connected, and, and similar things. So that creates files. So that device execution can be turned off, by users, by the admins in case you want it. But, we see, like, a lot of productivity gains, but if you want to take it gradually, you you definitely can. And the last piece is local MCP servers. So, really depends on your architecture model, how much of your firm is like self developing stuff, but we're seeing, at least the bigger ones use, leverage a good amount of local MCP servers that can be blocked, that can be, stay open, and, and you can definitely choose whatever you want. We don't think that there is right or wrong here. It really depends on how you architected your network and make sure that the core configuration fits into it. Perfect. And and just to reemphasize, these, are available in in Teams and enterprise for sure. We can check about Pro Max. I think it's in every Claude desktop deployment you have, but, but, definitely for Teams and Max. Incredible. We released in August, our compliance API, support and made it generally available in, in September, and that was a very big, big hit for a lot of, customers that I'm in touch with in the financial sector. I know at least one firm that, the following day after we released it, have done the test. They've been in touch with us in advance, and we gave them a heads up. The following day, they've done the test they needed. And the following day, they went firm wide with co work, because this was such a big hit in the material milestone. So the compliance API is our name for our audit log and the programmatic way to retrieve the top information, what's in the top and key information. MCP, the session transcripts, as you can see, like, everything that the end user, let's say, Drew Parker, asks as an end user a few questions, everything goes in and out from the model is being, registered there. It's the prompts, responses, toolbox. That's the vast majority of what our customers are looking for. It also allows activity fee when they drew log in, log out, similar things. It shows both what happened in the desktop version and in the cloud. It wasn't the case in early August. So, really, really, a a new innovation, and and things that we've delivered. And really sending that you can send and streamline the data to your SIEM and DLP. That's the vast majority, of of how customers use it today, especially in in this sector. So you can already straight with a 100 security integrations to to where you are. We used to have in the past, OpenTelemetry, OTEL, as you can see in the bottom. We're continuing to support that, but I would say that the compliance API has more extensive data, and you can see in our public docs the the comparison. And also the fact that from an architecture perspective, you don't need to connect every endpoint to your data, data lake or SIEM, and you can just pull it instead of get it, in an ephemeral way really allows customers a better architecture, and that's a big feedback, we're getting. So if you already have invested in hotel, you can continue using it. I would advise you to consider moving to compliance API. But, definitely, if you're in a a clean state situation, we would really advise to take a look on the compliance API and see and and get the full data and the future support that you need. Incredible. So that's, that's a lot of of what we shipped. Another thing... And I'm seeing a bit in the in the questions. Yeah. And compliance API is available only for enterprise teams, enterprise plans. We do understand the issue there, but it's it's part of a a single sign on is, there are few additional things for security and compliance that actually go to the enterprise plan, and we propose you being in touch with the accounting if that's, an issue. Connectors. So I'll say that a big piece of of moving, for a goal oriented AI, as Drew said earlier, is, connectors. Like, if you're just copy pasting emails, back and forth or fixing text, that's one thing. But once firms start adopting three to five at least connectors, they're seeing the material value. Because AI can now really accelerate you and and your... And, and employees with correlating data for you, which usually, is something that's pretty complicated and we need, like, a lot of human capacity for. With that in mind, connectors are really... Have done, like, major progress in the last few months where we really listen to admins and how you want to set it up. So few things as as guiding principles, they are correct not only for core, but I think really, like, becomes more important as you adopt core and move to goal oriented AI usage. So first of all, admins can decide and set the ceiling. So you need... The admins need to connect it, to configure it, and they can set in a very random way. Each connector has its own granularity, but I can say that I'm very close to a few connectors that, you use in a very granular way what end users cannot do and can do based on your threat model. Everything comes with a default, but you can you can edit it. For example, I know some firms that started without being able to, write, on on some, connectors, but as time evolved, they they allowed it. Anyways, for example, and even here in Anthropic, no mail is being sent without my approval. So so you can really, target and cater what needs, like, more permissions, what less as as you wish. From there, users need to opt in to every connector. That's the second piece. And they can also opt in, to the ceiling that you that you set, and they can be only stricter. For example, they can not allow every action, so on and so forth. I know that I personally haven't... I've taken a slightly more conservative approach in one of our connectors that I wanted slightly more control over because it was important for me. I wouldn't say it's the vast majority of what end users do, but it's definitely, possible. And then approvals, are really, really a good way. Nothing has happened automatically unless the end user, allowed it, prior. So approval on every time where you can have a very direct connection between what the AI is doing versus what the human, allowed. So there is no problem of ownership, who did what, and so on. And that applies to... Throughout our our, our product. They just, as I said, slightly more relevant for Cowork where users who connected more connectors really... Like, we just see... We see the impact and we see how they're referring to AI afterwards. And I know that maybe as as, if some of you haven't connected, more connectors, there's a bit of a chicken and egg that it's unclear why should it be helpful. But in the same time, you don't realize the full value unless you've seen it. So I've seen... That's part of why RBAC is here, but also you can look at at the industry. I've seen it very successful firms adopt dozens of connectors, and although some of them are internal MCPs. And, really, via RBAC, you can control certain populations, start small, and, and evolve over time to just even get convinced on your own, for... On the value that those connectors provide. But, really, that's a very material part of, AI, utilization, especially with end users who need to really correlate information as part of their job. Amazing. So with that in mind, and we'll share share this later, but, top security questions that that you have in mind. If there's anything here that you would love us to follow-up on, feel free to skim and feel free to open, to asking the questions, later. But definitely, really, really feel free to take a look and and, hit us with our questions later. Yeah. Drew, anything, before we go to q and a? Anything. you would like to add? 100%. I think one thing that might be helpful on for this audience to do, like, a quick double take on is a lot of these RIAs and financial advisory shops operate within teams. And so there are individual advisors or maybe a single advisor that has his... Their own book of business and then other teams that operate collaboratively, and share clients among themselves, but not with another team within the same firm. How... And this sort of lends itself to the, where does data live, and what, like, tenant, what are the... What's the overlap and risk of, contamination, etcetera. When that's the case, how can one adviser team be confident that their data lives within their team or set of instances or licenses, I should say, within Anthropic and not going to be shared with the rest of the firm and all the other advisers. Yeah. For sure. So, we've made it the approach definitely even if you look at at our customer base, it's not like the majority are are, like, the huge enterprises, like, potentially other software that has a huge sticker shock. Like, the software here is really, like, a lot of self serve, a lot of, like, the lower end of, like, dozens... Low dozens to high dozens to hundreds of of, end users. With that in mind, that's how we we designed it to make it low touch for admins. Obviously, admin... We we give them the balance to go in a t shape. If you go very broad, things are well. If you go... Wanna go very deep, you're allowed as well. Like, if you want to set a certain connector with, like, the end derivative of, of, compliance and, and, permissions scrutiny, you can definitely do it, but it comes with a fairly, secure, default configuration. So with that in mind, for example, on projects, unless, an end user in one of those teams share it explicitly, it's not shareable, and it's not sure... Shared by default as it stands. Likewise, with data and sessions, data is yours as an end user unless, unless you explicitly mentioned it and shared it with your team. And with that, like, that's how we design the system. So there is no cross contamination between end users and similar. So even if your Claude admin did nothing, we promise a a secure a secure experience because, for example, if we just go back to encryption keys, if you're, like, one of the biggest firms out there, you might have your own encryption vault and similar things. But if you're not, you're just getting our very strong, that's a FIPS compliant, encryption, mechanism that really allows by default simple execution on that. 100%. Yeah. Thanks, Dor. I think to sum up too for everybody, let's say Dor and I decided to start our own RIA, Claude for investment advice or whatever it is, and we, operate individually. We have our own books of business. What Dor inputs into his Claude, even though we're on the same license, I cannot see in my Claude. I... And myself and my Claude can only see what I have access to within the connectors that have I... I've set up and the individual sessions that I have, or as Dore alluded to, shared projects that we create to actually collaborate on. So user access level permissions persist over connectors. So if we connect in our financial planning tool and our portfolio management software and what have you, what my Claude and I can see in each of those tools is tied to my access level permissions. So, it is a really good question, but, honestly, the existing sort of out of the box state and config within the enterprise license and, honestly, the team plan too will suffice for that entire framework. But... Yeah. I think... Or anything else that you wanted to touch on quickly, or should we hop to, just some of the questions that have been coming in via the chat? Yeah. I think I'll I'll just take one of the questions already and and connect it to what you said. And likewise with PII, like, I think in in general, we have a lot of sensitivity to your clients' PII. We take it very seriously internally on who can even access the data and also at at the product level, resegregating. So user a doesn't see user b's data, because of those exact reasons. And we've really, like, designed it, by default in that case. A 100%. One thing I'd actually like to add there too. So, for the same reason that you can enter in the most sensitive client data, even Social Security numbers or anything else into the systems that you trust today is because they have the requisite info security controls. They're SOC two certified, ISO two seven zero one. They have all the things that you need to trust them, and you have data isolation. It is the same thing within Claude. I think, oftentimes, people, like we talked about earlier, will get confused on if they need zero data retention in order to enter in client PII or not. You do not. What you need is the data isolation piece, and the necessary, like, audit trails and other, like, enterprise compliance features to correctly monitor and archive everything where necessary. But, yeah, just to be very explicit, you do not need a zero data retention enterprise license in order to enter in client PII. And to take it even a step further, one, zero data retention is not available on the enterprise plan. Full stop. Doesn't matter if there's a minimum spend commit or anything. It is not available. And second, even if it were available, you wouldn't want it. If you had an enterprise license with zero data retention, you would be left with a Claude that is incredibly smart but not super useful because it would lack all the context and memory that it's built up over the duration of time that you use the tool, for an amount of months or eventually years, etcetera. And so data retention is is a good thing, inherently. It's also what helps us to run our classifiers and make sure that, like, nobody is taking over your Claude and running, like, prompts that are potentially malicious or problematic from, like, a bio or cyber perspective. None of that is possible without data retention. But, anyways, Yeah. And and just to add, if you want slightly more, like, written content about it, feel free to go to our trust center. So trustofontopic.com. It has a pretty extensive list of of guides and and explanations on how we secure PII and and, our compliance certifications that Drew mentioned, so on and so forth. So that... That's a great resource. Yeah. Awesome. Okay. We'll go through as many of these, questions dropped into the chat as possible. There are a lot, so bear with us. I'll try to handle the ones just from, like, a a business perspective, and then, Dor, I'll I'll tag you in for some of the security related ones. To start, how do sole proprietors and small firms, use Claude for financial advisors? What's the correct license? Lots of questions on that front. The best way and our recommendation to do this is to spin up an enterprise plan. We recognize that the average RIA does not have 20 people at it. I think it's something like eight people work at the average RIA. But the enterprise plan is still the best place to start with. The enterprise plan requires a minimum of 20 seats to be purchased. Seats cost $20 per month for the year, and that's billed upfront, for the year. So it's $4,800 in upfront costs to gain access to the enterprise license. If you don't have 20 people at your firm, you don't have to allocate every single one of those seats, to somebody. You can have unused seats. They won't incur any usage fees, etcetera. For context, the team plan is designed for small unregulated businesses that do not have the same archiving constraints that you need. And, info security, all of the questions that we talked about today are not going to be asked by a mom and pop shop looking to use Claude to optimize their food delivery system or what have you. So, I think the enterprise plan is the place to be regardless of firm size. For context, I am working to see if there is something that we can do to potentially lower the seat count for the enterprise plan. For the time being, there is no path to that. And so if you want access to the tool, you'll need to purchase the 20 seats to begin with. Longer term, we'll see what we can do. But moving. on. Yeah. Maybe to add one more thing because I'm talking to a few customers myself. Some people, like, I think a year ago, they couldn't self serve by it. And today, we've improved it, so you can even, like, upgrade at your own time and so on. And that's, you know, relatively new in, like, software 100%. Also, if you have an existing team plan, we can take. that. and upgrade it to enterprise. It's an in place upgrade like Dor's mentioning. So, no data settings, chat history, etcetera, get lost. Everything persists. It's a back end flip on our end. So pretty simple. Next up, let's see. When will eMoney be a connector? I think this applies to all other tools that you're using today that you don't see a connector available to. We very much intend to release additional, like, waves or launches, if you will, of the Claude for Financial Advisors plugin. We want to include as many of the tools that you're using. I think the the process for context is we have an MCP submission that we collect on our side where the third party tool or vendor submits the MCP server to us. We review and eventually approve if everything looks good. To the extent you want a tool included and have a connector... To have a connector in Claude, let them know and ask them to submit their MCP and work with us to to get it reflected in the Anthropic directory and embedded into the Claude for financial advisors tool. Continuing on. Does unselecting the help improve our AI models in privacy settings give a similar level of privacy to an enterprise subscription? I'll let Dor opine a little bit on this, but for context, yes, that will turn off training on all the inputs that you have into your individual pro or max plan. However, what it won't turn on is all the enterprise only features, like the compliance API, RBAC features, etcetera. But, Dor, if there's anything you wanna add there. Yeah. Exactly exactly what you said. And it slightly varies per per plan. So, I would advise you to really validate at your own plan what's what's that. 100%. Yeah. Next up, what features capabilities are most in demand from wealth managers? My current clients have predominantly identified a need to be able to orchestrate wealth specific workflows across multiple systems. I think, two two fronts to this. There there are both the front office, like the advisory and, like, end client facing workflows as well as internal back office workflows like the ones specified in the question such as onboarding, account opening, etcetera. I think we have, like, a v one sort of in place with the existing Claude for Financial Advisors plugins and the workflows that we ship there to the extent that back office workflows, like account opening, etcetera, are interesting, and you want to work with Claude on all of those, which, by the way, lots of firms are, give that feedback to the individual tools and systems or custodians that you want that workflow to be available. I think all of the partners, MCP partners, I should say, that we work with to set up these MCPs, they're very keen for your feedback on, hey. What actual fields or things would be helpful to surface in the MCP call that we set up? So that feedback, I think, is very wanted by the MCP partners. They're the ones that determine what is actually available for use both in a read only and an edit capacity over the MCP. So I think that's feedback that should absolutely be delivered to them. But the sort of, like, general thought of the question is a 100% accurate. There are absolutely firms that are tackling back office workflows, everything from onboarding account opening to a whole litany of other items. So, I I think it's it's a great place to to focus your time. Dor, this one's for you. So what subscription level of Claude do we need to access all of the features and info security stuff and local versus cloud, Cowork settings, on cloud? Yeah. So so the enterprise one is that is that subscription. And, to alleviate, other concerns of how maybe other software, how other software builds it, you don't need to add for additional secure... Like, the security controls are coming by default in that enterprise tier, and you don't need to add more with the sole exception of US inference only, which is relevant for only very specific in the sub industries in some some capabilities, and I'm not sure that this is the focus here. But, that would... Buying that would just allow you to consume everything we've discussed here, by default without any, like, concerns and things. 100%. Next up, our team is curious about the cloud team versus cloud enterprise security layers for client PII and NPI. We're currently on cloud teams and are SCC registered. Is Cowork still safe on the cloud team plan? Yeah. Is... I... Go ahead, Dor. Yeah. Yeah. I I would say it saved the question that follows is how much control and granularity and auditability do you have, as you use it. So it's slightly more like a question for your, auditors as well. But if you do need to prove to them what happened in every prompt and response as part of it, then that's available in the enterprise tier, but not in the team's tier as, as you mentioned, Drew. Yeah. 100%. So is Cowork safe to use? It... Does it have all the same, setup features, etcetera, that we talked in-depth about today on the team plan? Yes. What does it lack? The enterprise controls that we went over as well. So, for example, if I'm an RIA, what am I thinking about in deploying any LLM and agentic capabilities? I, one, I need an AI policy that is embedded into my disclosures. Two, I have a compliance API set up so that I can archive all of the chat history sessions, actions that an agent takes on my behalf, etcetera, into my archiving tool. And, c, I have RBAC controls set up so that everybody can only access what they need to. And I have spend caps put in place for individual users or groups of users, etcetera. All of those are only available on the enterprise plan, which is why we are prescribing it or recommending it to RIAs and other regulated customers given the environment that they exist in. Dor, why are you transitioning from local to cloud? Yeah. So I think in general, the the main values that we're seeing are end user values and and security values. I think especially as we've discussed with a lot of people here, you would love to consume as secure by default environment, as you consume. So from an end user perspective, you can really now leverage, your web browser, mobile phone, and, and also really allow longer tasks while you were, just not there, with your laptop. That's one on the end users. On security, I would say that, we're putting a lot of effort internally. There are, like, hundreds of people are working on securing, our customers' adoption. And by moving to cloud, we then reduce the dependency on what security you put on your laptops and endpoints. With that in mind, that really allows centralized governance and additional values that really isolate your customers, your clients' data from, from, the laptop and similar and allows a better security and easier security setup. 100%. Thanks, Dor. I'm just seeing a lot of questions on PII entering it into the enterprise license. Is it safe? I think to just give some additional context on that, like, what does what do I mean when I say... And what does Dor mean when we say data isolation? Essentially, what you enter into your cloud is specific to your instance. When you enter in sensitive client information or any other information for that matter into your cloud license, that stays specific to both the individual user and that tenant as a whole. Nobody from Anthropic can see that data. We can't package it up and sell it to a third party. We can't train our future models on any of it, and that is contractually... We're contractually precluded from doing so as outlined by the commercial terms of service. So full stop. Yes. You can enter in client PII into your enterprise license. Continuing on. Dor, this is a good one. Can you clarify the ninety day retention rule? If chats exist longer than ninety days, is that data retained? And to build on top of that, is it possible to have more granular controls on data retention such that chats can persist for an amount of time whereas attachments can persist for y amount of time. Can you elaborate there? Yeah. For sure. So we differentiate two main things that we heard from customers. One is chats that are usually personal, and second is projects that are either personal or shared as as you wish. So you can apply different, retention policies to to either. So less about the attachments, more about the method of work, where you work because then if we take just an attachment, but keep the chat, the the experience is is sometimes flaky. With that in mind, how it's set up is that, if you set up custom data retention, you can choose the minimum is thirty days, but then you can choose how much you wanna set. 90 is just a median, the dancing across, across customers. And, we really advise, using it in in the granularity that you want. I would suggest if you're, not, let's say, like, few thousand people, in in your firm, keeping it simple is definitely my advice. So using, for example, ninety days for everything is is what I would choose. And from a mechanics of how it actually works. So let's say, on the first day of the quarter, you started the chat. Unless you've... If you haven't touched the entire quarter, it's gonna be deleted in the end of the quarter, if you choose ninety days. But then let's say if mid quarter, you you open that chat again, you edit a few more prompts, then the ninety days, if you set it for ninety days, are counted from the mid of that quarter, not from the beginning of the quarter. So for as long as the chat is used, it... It's kept alive. So we keep only data that is relevant and highly valuable for you rather than state data that is just open and exposed. Like, obviously, security in a very secure way, but just adds more exposure to that we were trying to minimize. This is a best practice more than anything. Love it. Next up, due to regulatory constraints, I need to be able to reconstruct how AI assisted analyses were performed, including the prompts used, the outputs generated, and to extrapolate a bit clogs the reasoning. What level of information on this is captured in the compliance API versus hotel versus other systems? What granularity do we have access to? Yeah. So the compliance API is very granular and has everything that that, you just said in a very long, like, even if prompts are very long and and similar, and that's the the approach we advise. OpenTelemetry OTIL has somewhat the same data. Sometimes it trunk truncated. Sometimes you need to filter and to add a few flags for the full data to flow. But in general, we really advise focusing you on the compliance API for this exact task because it's also less volatile. So it's... If there was, like, a network hiccup on your side or anything like that or a laptop got turned off, or crashed, like, then you miss some of it. So, really, the compliance API is the most robust, way forward. 100%. Well, with that, we just wanna thank everybody for joining. We're a minute over here, but we really appreciate the time. These were excellent questions. And, yeah, I'm looking forward to to seeing what you all build with Cowork.