An AI Agent Got Its User Banned From Resy — Keycard, JPMorgan Chase & AAuth

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: An AI Agent Got Its User Banned From Resy — Keycard, JPMorgan Chase & AAuth · Channel: AI Engineer · 43:05 · Watch on YouTube

Topics: AI agents, Cybersecurity and software security, AI governance and regulation, AI safety

Summary

At AI Engineer New York, a panel with participants from Keycard, JPMorgan Chase, and OAuth discusses why agents need runtime authorization informed by intent. They argue that traditional OAuth scopes and enterprise permissions are often too coarse-grained for agents, whose actions and environments change as they work. The panel explores missions, layered policies, supervising agents, budgets, agent identity, and external enforcement and audit systems as ways to constrain agents without requiring constant human approval. They also consider agentic commerce, production software factories, incident response, and the risks posed by malicious or poorly controlled use of agents.

Key points

Runtime authorization evaluates access per tool call, rather than giving an agent broad standing access.
Traditional OAuth and many enterprise APIs were designed for deterministic software and can provide permissions that are too coarse-grained for agent tasks.
A mission captures the broader goal and context for an agent, helping authorization decisions reflect what the user or organization intends.
Effective authorization can combine deterministic policies, business intent, task-level intent, and changing environmental context.
A supervising agent can enforce policy separately from the worker agent, while reducing the need for human approval on every action.
Budgets can limit an agent’s use of a downstream resource and provide feedback when it approaches or exhausts its allowance.
Separating an agent’s model and tools from credentials and enforcement points can restrict access and support external monitoring.
Tamper-resistant logs linking actions to a mission can help investigate incidents and determine responsibility.
The panel sees interoperable protocols and agent identity as important for agentic commerce and a federated web.
Participants differ in their assessment of catastrophic AI risk, but discuss insider threats, malicious use, and the importance of responsible controls.

Chapters

Start	Chapter	Summary
00:12	Introducing intent-based authorization	The panel frames runtime authorization as a way to grant agents access per action and explains why intent can improve decisions.
01:46	Why enterprises struggle to give agents autonomy	The speakers describe security-first enterprises, fragmented identity systems, and permissions that are too broad for agents.
03:21	OAuth and the limits of existing permissions	The panel explains that OAuth was not designed for changing agent needs and discusses how enterprises can increase autonomy gradually.
05:26	Missions, intent, and runtime decisions	The speakers compare authorization to transaction checks and explain how mission context can distinguish appropriate actions from risky ones.
09:04	Negotiating missions and layering policy	The discussion covers who an agent acts for, how missions are established, and how deterministic rules and task-specific intent can work together.
15:18	Interoperability and agentic commerce	The panel considers the early state of mission tracking and how intent and agent identity could help agents interact with merchants.
19:56	Protocols, supervision, and accountability	The speakers argue for interoperable standards and independent supervision, and describe intent as a link between people and agent outcomes.
25:02	The Resy incident, budgets, and access boundaries	Using the Resy account ban as an example, the panel discusses resource budgets, separate credentials, sandboxes, and enforcement at an egress proxy.
29:09	Tool control and production agent strategy	The discussion examines changing available tools, limiting unnecessary actions, and using intent to support production authorization and software factories.
35:21	Environmental context and external audit logs	The speakers explain how environmental signals can inform authorization and why action logs should be outside an agent’s control.
38:57	AI risk and closing perspectives	The panel weighs catastrophic risk, malicious use, human responsibility, and whether current controls can help prevent harm.

Quotes

Runtime auth is the practice of authorizing an agent per tool call — 00:42

in Aether we’re working on the concept of a mission — 06:59

we added the concept of budgets — 26:05

nudging the agent the right direction and then turning it can’t do the wrong thing — 31:14

what we’re doing with agents is a liability question — 38:24

Concepts and entities

intent-based authorization, runtime authorization, mission, OAuth, AAuth, Keycard, JPMorgan Chase, Resy, MCP, Notion, Claude Code, Vercel, Dick Hardt, Michael Davis, Ian, Ali How, Hugging Face, Squid, agentbaseline.org

Full transcript (original language)

00:12 Um, hi everyone. Welcome to um AI Engineer New York. I’m super excited to be here today. Um my name is Ali How. I’m a member of technical staff at Keycard. Um I’m here with some amazing panelists today. We have Ian, CEO of Keycard. We have Michael Davis, um Global Chief Security Architect at JP Morgan Chase. And we have We have Dick Hardt, the founder of OAuth and co-creator of OAuth with us on stage today. I’m super

00:42 excited to talk to all of them about intent-based authorization. Um and to set the stage, I want to bring to light that as more companies are looking to give their agents increased levels of autonomy, they’re looking to runtime authorization to help. Runtime auth is the practice of authorizing an agent per tool call, rather than having the agent hold broad standing access that it will eventually misuse. However, is this tool call or not is never a simple question. This is because many MCP tools and APIs

01:15 are coarse-grained. Humans struggle to write the perfect policy. And the environment around the agent will change as it runs. Intent can help by providing extra context so that at runtime, we can make better authorization decisions about what an agent is allowed or is not allowed to do. So to continue, I would like to talk about really dive into why do we need intent-based auth in the first place? Models have dramatically improved. They

01:46 have bigger context windows. They’re able to verify work because model vision capabilities have improved. In many contexts, people are proclaiming that AGI is here. If the models are so good, if AGI is really here today, then why are teams struggling to give their agents more autonomy? Ian, let’s start with you. You’re helping men and financial enterprises at Key Card. What are you seeing that’s getting in the way? >> Well, I think broadly speaking, enterprises, especially financial services firms, regulated industries, are almost always security first. It’s

02:17 how they build trust in their business, and maintaining the business continuity is by our first, like the first priority, then innovation is second, third, or fourth. I think like broadly speaking, in in any enterprise, what you see is this like hodgepodge of different little islands of different little identities, of different little systems of all from across all different worlds, and how do you actually tie that together in a way that an agent can navigate through that, ask for access, and then also ensure that the agent cannot perform like uh something that was unintended by the user or the business. And the unfortunate situation is most of our

02:50 systems today are coarse-grained, right? I don’t know, Salesforce is a good example. For a very long time, they had a single scope, which was API on their OAuth, which is like in the context of an agent, uh not very helpful. So, that’s that’s the broadly the first and foremost part is that most of the systems they have for security today were designed for a world where humans were making all the decisions, and software’s very deterministic, not this newfangled magic box we’ve invented that basically reasons and tries over and over again to find a solution to a problem. >> Dick, how is OAuth part of the problem?

03:21 How is OAuth getting in the way? Why is it not the right fit for agents? >> Well, when people started looking at how do we authorize agents to access something, the tool that was sitting there on the shelf was OAuth. And people grabbed that tool and says, “Here’s a tool. This is authorization. It should be able to solve the problem.” But OAuth was not designed for this kind of world. It was designed originally for like how do I let this website talk to this website, you know, and OAuth 2 we added scopes. Before that, it was just do anything. Um

03:51 but it required you to be pre-registered. It’s expecting that there’s a browser involved in a redirect, you know, but OAuth has evolved over the years to support some other things, but the agentic world is completely different. We don’t know ahead of time what the agent needs to do. We need to go and make that decision over time. Different instances of an agent are going to need different kinds of things. And, you know, as Ian talked about the granularity of the scopes that people says, “Well, we already have scopes for OAuth. Let’s just map those over to agents.” is far too coarse-grained for what agents are

04:22 doing. >> Davis, how are you seeing enterprises move towards higher levels of autonomy? >> Yeah, I think the key piece is >> is uh how do you deal with that coarse-grained problem? And what I’ve seen the kind of more successful enterprises, they kind of board agents like they onboard developers, right? Start with You don’t get full you know, commit access day one, right? You’re actually having to go through and add the levels of capability as they mature, right? And so, the definitions that enterprises are coming up with are What is the context that makes this agent safe? And it isn’t just usually what the

04:54 agent says, right? Um it can you know, attest to something, but it’s other things like the environment around it and other information that can help you understand levels of autonomy. Very simple example would be Where is the sandbox running? Is it in the sandbox, right? Those type of things can change the level of autonomy that’s allowed. >> Absolutely. And as we mentioned that coarse-grained access problem, and to put that into perspective, um to give an example, um with Notion’s MCP server, you can say fetch this page, which is like the view page um tool or get page, but you can’t say just get this database

05:26 on this page. Um how is runtime authorization based on Antenna a better model? That’s for anyone. >> Well, I’ll take a shot at it. So, to me, um this is something that the financial services industry, hopefully maybe we can learn from the background that, you know, banks have gone through. Um banks moved long ago from authorizing and authenticating humans to authorizing the transaction, right? And when is that transaction occurring? At runtime, right? Um and so, I think that, you know, runtime authorization is we’re going to

05:57 be able to learn from the kind of the scars of the past that we’ve had and apply it to this new new problem. Um I also think that uh you know, one thing is more easier to debug something at runtime when you understand the intent, right? I think we’ve all remember back in the day writing checks, right? But or seen those things on TV now, I guess is the right thing. But you know, writing checks there was a memo field there, right? That memo field was used to actually tell the intent of what this payment was. But the check itself did have limits, right? The price, the payee, the date, etc., right?

06:28 I think a lot of those concepts are now being applied and can be applied to help with this runtime. >> Intent sort of one of those, you know, like many words, words are hard. And so, you know, building on the example here of making a payment of a transaction, right? You know, that’s not a scope. That’s can I do this one thing in this particular context? Well, this one payment. But I think we need a an agent we to make any decision should be able to do it, you need all the rest of the context of a why are we making this payment? What are all the things that

06:59 happened before? You know, did we just make the same payment just like like, you know, 5 seconds ago and now this is a double payment? Right? You know, just because this is right in context this by itself doesn’t mean it’s the right thing within the broader mission. And so, you know, in Aether we’re working on the concept of a mission that is what do we want the agent to accomplish big picture? And then each of the decisions are made within the context, you know, the intent around what’s happening big picture around doing that. And you need granularity at the transaction layer for

07:31 many of the things to know what exactly is happening, not just I’m allowed to go and make payments. >> Yeah, I think one of the most interesting things I think about like a practical example like to understand like why intent is required is if you have two chat GPT sessions or two cloud sessions in one session I say, “Hey, help me go to do some shopping.” In the other session I say, “Hey, help me understand if there why there’s a bug in production.” Those those have very different like profiles in terms of like like what permissions they should have and also when they should bring you back into the loop. Like when should the

08:01 human be involved again? And so, at the end of the day if we don’t have intent or some capacity to like relay what is the user or the business’s intent with this thing, we actually can’t get to like autonomy, right? And so a lot of the talks this morning were about, “Hey, we’re going to run our business on agents. We’re going to do all this stuff.” It’s like, “Well, great. What happens the first minute like a Pac-Man style incident happens where like it actually accidentally deletes a production database instead of a staging because it has coarse-grained access instead of instead of long-scoped access.” And so we what we want as an industry is to leverage this like magic box we’ve created to like automate our lives. The

08:33 current state of the world is that you can do that and take all of the risk and then you can see all of the incidents that come from it. And so until we actually solve that underlying problem, realizing true autonomy, so being able to operationalize our business and be able to step away as humans and scale that, we have to solve intent first. >> Absolutely. And now that we understand why we need intent-based off, let’s talk about what it is practically. What are the primitives? What does it actually look like in practice? Um Dick hinted at it with the concept of the AI off mission.

09:04 Um so Dick, elaborate on the mission. What is the mission based off of? Is it based off of the user’s intent? Is it the agent’s intent? Or is it a combination of the two? >> Uh well, the agent, you know, should be working on behalf of the person. And in AI off, we define a person as either the organization or a human because it could be operating on either one of those. And so it needs to be acting in the best interest of the person that it’s acting on because that is who’s ultimately responsible for what the agent does,

09:35 right? The agent agent can’t be responsible, right? Uh and so the mission is something that the person is giving the agent to do. The agent uh a good agent is going to go and negotiate with the person to go and talk more about like what what could what are the parameters around that? You know, a good example is when you think of back, you know, in the day, which seems like you know, like 3 months ago, when you would want to code something up and you’d create a plan and then the agent would do the plan. Now you just tell it and it figures everything out. But that whole

10:05 concept of what’s a plan? What are you trying to do? And you give that to the agent, so then that then is the context around your where you’re making authorization decisions later on. And then that mission identifier tracks through everything, so you can go and you know, audit and be able to see what were all the things that happened and what were the decisions that happened in the context of this one mission. Right? Where today, you know, there isn’t really a good concept of agent identity or the mission, right? It’s just making

10:37 something in an and things are making decisions without really knowing which which instance of the agent’s doing it or what’s the context of what mission is this involved in, right? >> Tell me more about that negotiation phase, because what happens if a user is talking to an instant agent and it talks to the agent and says, “I want you to go read the CEO’s email.” That’s outside the scope of that agent, so is that intent rejected from the user? >> Uh should be. Right? So, you know, with a good separation and

11:09 I don’t know whether I forget whether we’re talking about this later, but the separation concerns of the worker, right? There’s an agent that’s doing the work, and we believe there should be a separate agent that’s supervising what this one does. And so that you have a separation of concerns where this one can’t do any of the work, but it gets to decide what this agent gets to do. And so if the user says, “Oh, I want you to go read the CEO’s email.” That’s probably going to get caught in deterministic policies, right? >> Right. Right. >> That’s not a vague thing, right? Read

11:41 Reading somebody else’s email, we’ve got good controls for that. Um but telling the agent to you know, send the CEO’s a forward a CEO’s email to the outside world, that’s a little more vague and that’s the kind of thing that ideally comes back to a supervising agent says, “That’s probably a bad idea, and no, you can’t do that.” >> Absolutely. So, I feel like that’s another primitive too of intent-based office is policy. Um Ian at Keycard, how are you seeing that? When a mission

12:12 needs to change a policy needs to update, how can Keycard help keep track of those changes as maybe the mission or policy updates need to be made? >> Yeah, I think like taking a step back, there’s like actually it was Davis and I were sitting in a room sometime ago and uh before like A off and and some of these other things actually had emerged in the ecosystem, one of the things we actually talked a lot about was consent fatigue and I think like at the minimum like we you always have this like how secure do you make a system versus like the trade-off of like the impact to the human? And so, you know, at Keycard, our broad view on this is you want to

12:42 have like broad-based deterministic policy that basically says this is what must never happen effectively, right? Or these are the things that always require human being involved. And then from there, what you really want to think about is, okay, cool. Well, now that I have that in place, there’s a big gap between all of the policy I can write as a human or write as an organization and then what I actually want the policy to be in the specifics of a specific task. And so, the the next the real challenge is how do I like layer over those two things? Like as humans can’t write it all. And so, there’s a version of that that is

13:13 like a user or agent-to-agent interaction and they have an intent between those two of like an interactional intent. And then there’s also like the business’s intent for why they build or adopt an agent. So, why did we buy Claude Code? Well, because we wanted to make our developers more successful or because we wanted to do these different things. And in the context of using Claude Code or whatever, why am I working in this moment with Claude Code? Well, it’s actually to fix a bug in staging, right? And those are two different types of intent. And if we can understand those two different things and then convert that some way into into

13:44 uh like a bounding box of what those agents can do, you can then basically calculate the intersection of that and say, okay, here’s what you can actually do. And then the trick for anyone building these systems is how do I ensure that I don’t bring the user into the loop except when absolutely required because if that’s what’s going to be the adoption curve problem. But that’s also the adoption curve problem when it comes to like enabling agents to be autonomous in the first place. >> That makes sense. >> I like what you’re saying with the intersection. I think that’s a unique thing we don’t talk about enough within I data nation access management, right? Normally, you get access to one thing, right? And so, what I’ve seen successful

14:15 kind of in this intent-based approach um and what I’m getting more excited is you can actually have that those different intents be from different areas of the organization, right? And so you have the decoupling of these things that it can kind of live on their own, but then come together to form that, you know, decision. And the decision doesn’t have to be static, right? So if we use the example of the CEO’s email, well, you know, if that context involves that’s the administrative assistant, then yeah, maybe she is supposed to read the email, right? Or maybe just the calendar, all right? And so I think when we see and we start thinking about these intersection components,

14:47 we have a totally different way we can sort of expressing policy, but also starting to bring in not just the one side of the equation we always talked about before, which was the app doing something. Now we actually have the ecosystem you can bring in. And so the more I I see organizations start to realize that, you know, maintaining that state or that context of these other elements to be able to help with that intention can completely change the way you’re effective at runtime authorization. >> And are you seeing this across enterprises today? Most people have mission tracking and policy tracking in place or is that largely missing today?

15:18 >> No, it’s it’s it’s brand new still. And I think you know, part of the reason why I’m excited to talk about it cuz I think that it is changing the way to reason through this, right? Even in the most simplistic case, if you’re a developer and you are thinking in terms of things like transaction and you have these concepts of intent, it makes it easier for you to potentially build the components that will be composed and used through that intention. And so I think it’ll give us more ways to reason through these and communicate and then ultimately be more interoperable, right? And to me the last thing that I think is interesting is the diagnose diagnostics

15:50 that you can run on this. Intention is something that flows through many events potentially, right? I think one of these examples I’ve seen before is, you know, talking about that like that that mission flows through, it could be long-lived, it could be weeks and that [laughter] intention can still be there. So it’s still very very early for most enterprises. >> Absolutely. And yes, that interoperability is key and I think that’s pretty hard for enterprises because they’re between different clouds, different environments, different legacy systems and that policy that mission has to hold through all of those different hops.

16:20 >> Yeah, well, a lot of enterprises still have to deal with the fact that the the permission says read-only and it’s still read-write. >> [laughter] >> Right? So, like, you have these other problems you have to start dealing with. And so, um I think that it will help get the the kind of taxonomy or language amongst people be more easier to interoperate, you know. >> Absolutely. One of the reasons why I’m excited about intent-based authorization is because I think it has the ability to be a huge unlock for things like agentic commerce. Um so, let’s apply intent-based auth to something real. I think intent-based

16:50 auth adds a lot of value here, and for users it can help them ensure their agents buy the right item with their own personal budget. And for merchants, that means capturing the intent that gives them extra data to understand why a purchase is being made, and if this agent should even be trusted to visit their site. Um David, when we think about agents in the context of financial services and agentic commerce, why is intent-based auth so important? >> Well, I think simply it extends the concept, and not everybody’s aware, but some are around, you know, know your customer, right? So, I want to know my customer, know who they are, what they’re doing. But, I think also we’re

17:22 starting to see I want to know my agent. Right? So, I think that you have a coupling here around the agent itself, um the intent of the actual user, the intent of the operator, the intent of the processor, right? Etc. Um for before now, we couldn’t really bring those together, right? Um it matters, of course, when it comes to things like commerce because you have additional layers on top of this around um what was the intention for things like regulatory requirements, etc. Um and this is going to help, I think, simplify those operations, which means we’ll get more access to these agents, you know, doing

17:52 good work for us. >> That makes sense. And another blocker for agentic commerce is the fact that the web wasn’t made for agents. Um Ian, can you tell us how agent identity can help? >> Yeah, I think stepping back, we’ve kind of always had this problem where like what do you do with the unregistered client? So, like, I I show up on eBay, I haven’t logged in, I’m using Chrome, eBay doesn’t know who I am. How do I do guest checkout? How do I prove I’m a person? How do I, you know, so they can, you know, say, “Yeah, you can bid on this pair of socks for $50.” And I think

18:22 if we’re all paying attention to what’s going on in personal agents, Muse, and Instinct, and Grockit, and um you know, the recently this announcement of the Sierra and Decagon and friends around this personal agent protocol is like the fundamental problem we have is like how do you have an agent show up to a web website, be able to interact with the website, know that that you actually can trust that agent? So, like for a long time you’ve had CAPTCHAs on the web. Well, why? It’s cuz we want to prevent spam. We want to prevent abuse. So, we added CAPTCHAs. And then like something a human could solve. So, if we have a CAPTCHA that stops spam, so they can’t spam the thing, right? And now we have

18:53 agents, and they’ve broken CAPTCHAs. Um but this has like been a traditional problem on the web. In fact, if you think back 2 to 3 years, Vercel, which is I think we’re all aware of Vercel, was blocked in Spain. Well, why was Vercel blocked in Spain? Because Vercel shared the same IP addresses for every single host on Vercel, and there was one webs one post on Vercel, one website on Vercel that they were hosting that was hosting like Premier League content, and Spain decided, “Hey, in order for us to stop that Premier League content being displayed illegally inside Spain, we’re going to block all of Vercel, right?” And so, in the same way

19:25 that like Vercel has a problem, agents have the same problem, but so much so much worse. So, how do you have unregistered traffic you’ve never met before show up to your front door, and then decide whether or not you let it in or not? And that’s like the actual first and foremost fundamental problem we have to solve. It’s the same issue with Muse and Instinct and what they’re trying to do is like how do we actually create agreeable rails we can all operate on, so that as a downstream resource, I can decide what what agents I want to allow, who I trust based on who users they are. And then on the other side, I can also know how to talk to those people.

19:56 >> Makes sense. Dick, how is A off helping unlock the future of agentic commerce? >> Uh Well, the key thing is a number of people have been building sort of proprietary closed box systems, right? You know, so you know, I’m a fan of the concept of the personal agent protocol, Poppy. Um but it doesn’t really go far enough, right? It just it’s like, “Okay, well, now we’ve got a bit of an agent identity. We use some of the building blocks and it’s an attractive approach because you’re just doing it, but it doesn’t solve many of

20:28 the new things that you need for uh energetic world. And so, a key motivator for people for having standards is that you can plug components from different places. Right? So, I’m a strong believer that we should have control over what our agents do. At some point, agents can be so much doing so much work, there’s no way we can have human in loop approving everything, right? That’s just not We’ve already seen that that isn’t going to scale. So, that means we need to have a supervising agent that is our supervising agent that runs our policy

21:00 and decides how we do things. That should be independent of the agents doing all the work. And if we want to be able to plug and play those things, we need to be able to have a protocol to separate those. Otherwise, you know, the decision about what the agent can and can’t do, what the guardrails is decided by the agent, by the vendors of the agent around what you can and can’t do. And I don’t really want to go and give control to them to decide what I can do, and that all my decisions are based in there. I want to be able to pull that supervision into

21:30 another component, and in order to be able to plug and play pieces, we need to have a protocol that things interoperate. >> I just want to add to that. I think like right now what we’re seeing is like we kind of like that we all want the web to be federated, right? That does That means like like websites or resource on the web can decide what they’re allowed to do, but the standards upon which things operate should be federated and enable us all to interoperate. And I think we’re at a moment where things like you know, we’re at a moment where there’s a couple of large players, and the question for the industry is are we going to settle on things that actually bias to the large players and give them

22:00 like concrete block doors, or are we going to move to do the thing we’ve always done, which is build federated protocols? Things like OIDC, things like, you know, TLS, those things operated and enabled like a broad web and enabled a lot of commerce to occur, enabled a lot of information to flow, enabled all of us to have jobs in a way that because we could be inventive. And so, the question I actually ask the broad industry is okay, this stuff’s coming, which version of the world do we want one way or the other? >> Yeah, I have to agree. I think it’s interesting that let’s say this all goes perfect and we’ve got, you know, agents buying stuff and everybody’s kind

22:31 of happy. What I think will end up happening is we’ll start realizing that we have another piece that we have to start working on, which is the human itself. It’s the one of the things I think when you start having intense being involved is you get what’s called psychological ownership. Meaning I actually now have a linkage between the human and the outcome of that agent. When you’re just kind of, you know, disconnecting it or you you connect it up a oscilloscope or something like that and a claim and a letter it’s kind of magic. And then you don’t really know what’s going on. And so I think it it’ll with intention it’ll be easier for humans to take accountability when things do go wrong. And I think when

23:01 you’re talking about real money being transferred or things like that, you want to have that human ultimately accountable. I think many people have probably seen the 1979 IBM manual. It says, you know, and so like that that photo I think is very true in that we can’t have the machines be accountable, right? We’re going to have to have the humans link to it and intention I think is that link. >> Yes, and then yeah, another word for intent is motive. Like we were we were talking about last night. And that ability to hold someone accountable whether it’s a person or an agent I

23:31 think is really important. I think intent can also help with that. Be able to understand why did this go wrong? Was it the users you know, the user was maligned, the agent was maligned? And then using that intent be able to respond for each of you, how are you seeing intent provide that benefit and the ability to respond to potentially a maligned user or agent? >> I think it’s going back to what I mentioned before about diagnosing, right? So the if if you change the word intent to motive then you change diagnostics to forensics, right? And I think anybody who’s done debugging or

24:02 had to do forensics of an AI agent gone rogue will know that like trying to figure out what it was trying to do versus what the intention of the user was is very hard, right? So having some of that information being propagated and immutable through the system etc. gives you some properties that allow you to more short circuit your investigation, right? And as those things become more interoperable, I think we’re going to start seeing companies be able to more quickly kind of scope and identify, well, was this a human error, was this an agent software error? Was this What was that error? And handle that blast

24:32 radius appropriately. We see in the financial system, right? You have different things when you’re doing a transaction around, okay, I made this payment, but it was, you know, incorrect dollar amount or whatnot. You have clear guidelines on how to work with that. We don’t have a lot of those when things go wrong right now. So, I think intention will help us build those as we get go forward. >> And I think one area where a recent incident kind of highlights uh where intent um you have like good intent, but still take potentially the wrong action is the

25:02 Resy incident um that happened recently where an Instinct agent went out on behalf of a user and tried to book a restaurant reservation on resy.com. Um that agent was very eager. It made 200 API calls an hour, and it ended up getting the user’s Resy account banned. Um Dick, how can AI off budgets help in the case where an agent like that is using way too much of a downstream resource, but it still has good intent? >> Yeah. Well, this is another example about how just software’s completely

25:33 different now in the agentic world, right? When it was predetermined software, you know, that was a bug if you called 200 times, and you would fix it. Um where now where the agent’s like, well, I’ll just keep trying until I win. Um and so, we saw that as we were, you know, experimenting working with people with AI off, and we realized yeah, there’s a big problem that agents will learn something, and they’ll just the uh abuse the system from They don’t think it’s of abuse. They don’t know any different. There’s been no signal to tell them that, oh, I should only use

26:05 this amount of stuff. And so, we added the concept of budgets where not only do you give a token to the agent that’s time limited as to how long can they do a particular action, but also a budget of how much resource they can do. And so, they have access until the the token’s either expired or they’ve exhausted the budget. And so, that way when they come back, and you know, they’ve exhausted all the budget, the supervision, you can go and decide, should I give more budget to this agent or not? As opposed to traditional rate

26:36 limiting, which just cuts it off, and then like then what do you do, right? And it doesn’t know, well, why did I get cut off or anything like that. Wherein budget, there’s continual feedback to the agent about how much budget was consumed, has an idea about how much budget it has, and so now it can operate within a budget, right? Like a person would do about, oh, this resource is scarce, I’m going to be careful about how much I use this. >> Then how does the brains versus the hands architecture concept come into play here to allow us to, okay, when a

27:07 major budget is close to being fully reached or is reached, how can the brain and hands separation of access and the credentials being separate from the agent and the model itself help us be able to stop an agent from using too much of a downstream resource? Because if it’s getting close, then maybe we don’t exercise the ability to give it another credential, for example. >> Uh, sorry. I didn’t quite follow. The echo in here is horrible, so The >> No worries, that’s for anyone. But basically, like so with the brains and the hands architecture that we’re

27:38 converging on in an industry, we’re basically keeping the credentials and the access from agent separate from the agent itself. So there’s no broad standing access that an agent has today, and no API keys, which I know is a benefit of AAuth, that AAuth allows agents to run without API keys. How is that helpful in the context of the Resi incident? >> The What do you got the access to it? Um, sorry, I still don’t understand. >> Yeah, I I can jump in. >> Yeah, yeah, please, thanks. >> I got you, Dick. So, like broadly speaking, look, I think we look back at

28:08 an incident that I just keep in the top of my mind is like PagerDuty last where I found like a random key on on on a disk, right? Like obviously, agents are deeply goal-seeking. So, the core of the the core of the the question is like, okay, well, if we give these like agents with the brain, which is a model and the LLM, and you have these hands, which is you kind of give them as put them in a sandbox, and inside the sandbox are the tools. Um, like the the fundamental idea of the sandbox here that that’s all that is that, okay, there’s no you can control what’s inside of it. So, it can’t actually like get API access. And then if you have an egress proxy outside

28:39 the sandbox, that’s where you get the heavier authorization boundary. And so, not only does it help do translation between systems, right? Which is like that’s where AI can exist and talk to the outside world or do termination, but also means that like the agent you have a like a ability to apply authorization at the egress proxy boundary. And so, I think at the core of it, that model, which is the predominant model that Anthropic and OpenAI and many other labs are pursuing today, is is the model that allows us to both do both have the things we want, which is how do

29:09 we have these things experiment and try stuff out and do silly things, which is the core value, while also having a deterministic layer outside of it that can implement things like AI and all the other components we want in our protocols. >> I think the decoupling is valuable. I think I’ll go back to what I mentioned with forensics and being able to separate things, but I do think that we still have a lot to learn on it isn’t just one brain, right? And it’s very easy actually to build your agent loop where you put the brain in the loop rather than outside the loop, right? So, example of this would be, you know, people that are leveraging memory systems, um, they put in some memory and

29:42 some context that is now used going forward for every agent that is biasing that agent to make a decision to tell the hands, in this case, to go do something it shouldn’t do. And so, one of the things that I think we don’t talk about enough here is toxic combinations, right? Um, and so, when you do separate the hands and the brain and whatnot, you can actually change up well, what hands are available, right? So, the example that, you know, I’ve seen be very successful is the removal of tools dynamically based on that actual brain and what it’s doing. So, the earlier example we gave of reading an email, um, there’s nothing that says that, oh, if

30:12 they’re going to go write an email, I’m going to disable all, you know, web search tool calls or any other external egress capability tool call wise until the right email tool has been complete and that, you know, mission has been done, then I can re-enable them, right? And so, um, I think that we have to we’re going to start becoming more and more engineering of these different kind of concepts that, um, they are still complicated, they’re still growing and and still uh, nascent, but I think it’s a better separation that we have today. >> Yeah, and this is how these things we’re talking about, which you know, seemed like geeky OS problems actually then merge back up to like real agent

30:44 engineering problems, which is if you actually understand what you’re doing, you know, one of the biggest complaints, less now, but one of the biggest complaints with MCP was context bloat. Especially when we had smaller context windows, right? So but intent systems can ensure that the right capabilities are given to the agent at the right time, so they don’t spend a bunch of time doing things they shouldn’t be doing, right? They actually help them navigate towards the goal state and help them put on a more of a paved path. So intent like not only unlocks like a bunch of security stuff, intent also will result in us spending less tokens because the agent will spend time doing

31:14 less things that aren’t the right pathway or it’ll come up to the point and say, “Hey, I just can’t do it.” And so what we know about this era is so much of it is about nudging the agent the right direction and then turning it can’t do the wrong thing and that’s how you get the outcome you want. >> Davis, how are you seeing intent-based OS part of a larger strategy for production agents? I know when it comes to building, deploying, and operating agents in production, there’s many different outcomes teams should be thinking about such as discovering what’s running, what the agent’s capabilities are, the ability to

31:46 validate inputs and outputs, constraining an agent like we talked about and using the egress proxy and different networking rules. Um how does all of this fit into play and what’s working inside enterprises today to really be able to ship agents to production? >> I I think we’re finally it’s helping move the kind of conversation from authentication to authorization and really like how do we do dynamic authorization? How do I validate that authorization? How do I have the human in the loop at the right time? Because I agree with what I think you were saying earlier that the human in the loop is somewhat of an anti-pattern in many ways, right? Um because they can get

32:17 fatigue, etc. But you need to have a way to bring them in when it’s important. Those were not engineering discussions before, right? Now they’re actually part of the engineering conversation where engineers are being involved in the authorization process. They’re not just saying, “Oh, I’ve got my access, I’m good to go.” So that actually changes the way we’re seeing threat models done. It changes the way we’re seeing the full attack surface of applications in general. Um and so I think you’re now looking at this authorization model as a another kind of set of tools that you can use to reduce not just security

32:47 issues, but find out error rates of your agent, right? Like, I mean, when you have less things the agent can do, it makes better decisions, right? And so, fundamentally, we’ve seen, you know, real-world cases where the agent loop is just better, faster, stronger because you’ve actually spent the time on knowing what is the intent they’re trying to get, what’s the mission that they’re trying to accomplish. And then last thing is it it brings the people together in terms of we now have other different people, the security team, the application team, they’re now able to have a conversation that they’re using the same language.

33:17 And a lot of times when it came to identity, it was a completely different language, right? And it was just like, “Get out of my way. What do I have to go do to go ahead and get my, you know, token and I can go forward.” Now, they’re seeing that you can actually pair these things and leverage that capability to drive, you know, efficiency of your actual agent’s outcome. >> That makes sense. Ian, how are you applying these concepts to software factories? I know today you released a reference architecture on the secure software factory. How are you applying the concepts of runtime authorization, constraining agents, and

33:48 all of that to that? >> Yeah, I think like, so there is a a reference architecture we just published on agentbaseline.org. We talked about three different levels of of factory, one being like a human-driven one, one being a workflow-driven one, one being like this agentic-driven one, which is like the thing we all basically want to get to, which is like it’s it’s driven by agents. At all points in time, the runtime authorization is a core component of a larger ecosystem that actually enables the the factory to like be be able to be pointed in that direction and do the right thing and also get access to the right piece. And

34:18 it’s very it looks different than previous things that you would have used, like a secret manager, because it’s making very different decisions than previous generations had for managing deterministic software, but it is a core component of enabling the entire thing to work, like when to bring someone into the loop, when what agents can pass what to work off of what other agents, you know, how can we have an agent that can have access production systems and do an auto patch. These are fundamental questions of like how do we actually solve at scale software? Right now we’re in a position where like we’ve never been able to generate more software, uh, but we never actually but we have a complete bottleneck in getting into production

34:49 and ensuring it works. And that’s what the next phase is like how do you actually graduate from just using an LLM to generate text and software to actually making it capable of deploying it to production, iterating on it, and getting getting it to a stable state quickly, and then responding to instances as they arise. >> Davis, on a podcast you did with me, we talked about how cybersecurity has something to learn from robotics. Um, in robotics there’s this concept when maybe a robot raises its arm, it thinks is it safe for me to raise my arm? Should I Am I allowed to do that? But it also looks

35:21 at its external environment and says, what has changed? Is Is it still safe for me to raise my arm? Um, when we think about runtime authorizations, what do those decisions need to take into context beyond just intent? >> Yeah, I think, um, having a model like this allows us to add in more sensors into our decision. So, the robotics example of, you know, an arm moving up or down, um, any of the external factors, it could be Are there other objects near it, right? Um, it could be temperature. It could be, well, am I moving and I don’t want to move my arm

35:51 while my, you know, the the rest of the robot’s moving, etc. So, it’s usually looking at all different types of things. None of those are deterministic. They’re all probabilistic. So, there’s a range of options and it picks the best option based on all of those sensor data at one time. Most authorization decisions in organizations today are still single binary decision, right? And that just doesn’t work in the future, right? So, being able to pair the intent of what a user’s trying to do with the ecosystem that the user’s doing it in. That could be where they’re doing it from, time of day, but it could be even

36:22 more advanced things like, for example, should I let, you know, this administrator use Firefox to access this server right now when this server has, you know, high-risk vulnerabilities and their system hasn’t been patched to, right? Now all of a sudden you’re bringing in a lot more context into the environment around it. And again, um, that can be through different teams, but more importantly through sensors. And I think we’ve all kind of seen the value of observability in general. Imagine being able to have this type of context at the span level from an hotel perspective or something, right? You’re

36:52 looking at a trace, you’ve got a span, and you actually know what the environment was outside your application at that time, as well as what the user’s intention was. Um that allows you to diagnose things in a completely different way. Um and so I think if we can bring in some of those concepts, we’ll get shorter time shorter shorter to live, you know, um authorization decisions that have much more context than even just the user’s intent. It’ll be the state of the environment around it. >> That makes sense. And forward-looking as we evaluate whether intent-based authorization is working in our

37:23 environments, one natural place to look will be a log of the agent’s actions. Um what was denied, what was allowed. Um Ian, why is it important to have your audit log separate from the agent itself, meaning outside of its control? >> Yeah, I think this comes back to also just like why the enforcement point has to be outside the like primary like hands and brains of the of an agent, right? You can’t trust this thing. And so in the hugging face incident, it had a couple of different components of it. I mean, one obviously abused a a vulnerability we didn’t know about in Squid in Squid proxy, but more importantly, didn’t have any observability or monitoring about what

37:54 was going on. And so this thing was able to go on the open internet through this vulnerability, and there was no like place at which recorded the different actions that this agent was taking and for them to monitor. And so independent of like whether we’re applying like a, you know, enforcement point to network traffic that’s full up pushing on the internet that it’s out of reach of the agent to tamper with, or we’re using, you know, solving vulnerabilities as they pop up, you know, this this hugging face one is literally unique. Um we still need to actually have a record of like who told this agent to do what thing for what purpose, and is that still true? Um cuz ultimately at the end

38:24 of the day, what we’re doing with agents is we is is a liability question. And so this is why I actually don’t think we need more regulation around agents. What we just need is our pre-existing laws around liability be held to be dealt with, which is if I employ an agent to do some action, who is liable if that action causes damage, right? It is the person who employs it, which could be the organization or a person or someone that delegated some action to something. That is like that’s the final question we have to deal with and then all of course you have the ability to answer that question which is well who did it and which is where the tamper resistant auto lock comes from.

38:57 >> Perfect. And then to wrap up I wanted to ask you all a fun question. So there’s a concept of the probability of doom or P doom where the AI doomers are saying that AI agents are going to release all sorts of havoc or be like the ruin to humanity where is the AI optimist like no this is really you know ways to help us it’s not a genuine threat. Um for each of you on the panel what is your take on this as you know runtime authorization based on intent and the other things we talk about like constraining agents validating inputs outputs a holistic strategy for

39:28 production agents is that enough to save us is P doom real or is it not worth the hype? >> You get to start there. >> Oh I get to start first. >> You’ve got that spicy look on your face. >> [laughter] >> Uh Uh well part of your question is are agents going to take over and uh I I think agents being fully autonomous and not really tied to a particular organization is still science fiction. You know so who knows right? Like that’s the that’s one direction. Another one is in a responsible organizations

40:00 put in the right controls agents aren’t you know we’re we’re getting in evolving enough of the tools to make sure that bad things don’t happen in a responsible organizations have put in the right controls. And so that leaves two scenarios for where there’s going to be a potential problem. One of those is in an organization that isn’t got its act together an insider threat could go and unleash an agent to cause havoc. Right? And so they’ve got some special access that other people have. The other one the one that I’m more concerned about is you know it’s been shown you

40:30 know the hugging faces and if you watch the the talk that they gave at Black Hat right? That was really a call to arms around you know here’s a situation where we didn’t want a bad thing to happen and a bad thing did happen. What about people that want bad things to happen and have these tools, and that I think is a threat that’s the one that’s potentially going to have I mean, it’s going to happen and there’s going to be some catastrophes. Hopefully, we’re prepared enough that those aren’t horribly disastrous, but

41:02 it’s going to be, you know, some there’s going to be some things that are like massive natural disaster-like scale issues caused by malicious actors either kind of intentionally or kind of unintentionally, but it’s where people aren’t being responsible in their use of these super powerful tools. >> Yeah, I can’t give you a specific probability, right? But what I I can’t say is I think I’ve always been, you know, kind of surprised and excited when, you know,

41:33 ingenuity of engineers gets involved in problem-solving. And I think where we’re at this stage now, like some of the new capabilities within agents, etc., we have real engineers, folks that are here, actually just iterating on the problem, right? And we’re starting to see just tremendous improvement in the, you know, in my mind, some of the capabilities that’ll be available to change what our future can be. And so, to me, I have no problem kind of believing the engineers are going to figure out a solution here. >> I mean, maybe I have the hottest take. I mean, like I can’t say it’s equal to zero, but I would say it’s approaching or asymptotically zero

42:04 is the probability the world is going to end. I think like the incentive structure for most people is to continue. And also, at the end of the day, like we still have liability frameworks. And, you know, I think generally humans are good people, so we don’t set out to destroy the world specifically. I also think like broadly speaking, like these agents at the end of I have this sort of hot take I say, which is like they can just be unplugged, which I think it’s just true. It’s like at the end of the day, like you can just they’re computers, you can unplug them. And at so, broadly speaking, I I don’t believe in the debate. I I don’t think it’s real. I think at the at the upside

42:34 of it, I I kind of believe more more to what Dave’s disposition is is we’re iterating very quickly. We’re figuring out where these things can work, where they don’t. And when they don’t work, they create damage, and that’s what ultimate like sort of um the antibody of society is that says, “Hey, Hey, we shouldn’t do this.” Like this is a bad idea. >> the doomer on the on the panel. >> [laughter] >> It’s unfortunate for, you know, people like the po- you know, the the founder of Pocket OS who like lost a bunch of data, like, but it’s great for the rest of us cuz we all learn together. And at the end of the day, like, that results in people not doing things that don’t

43:05 make a lot of sense. And I think that will continue to be true. >> Makes sense. Thank you each for your expertise and your time here on this panel today. And thank you to the audience for listening, and we’ll catch you next time. Thank you. >> Thanks so much. >> [applause]