Agent Identification

The internet has long ceased to be the domain of humans alone. Increasingly, programs are at work that carry out tasks independently: they search for flights, fill out forms, compare prices, or make purchases. Such programs are called agents. For the operator of a website or a bank account, this creates a problem: they see an access request, but don’t know who is behind it. Agent identification encompasses all procedures by which such a program identifies itself unambiguously. It should be recognizable which agent is accessing a system, on whose behalf it is acting, and what rights it has.

Why a password isn’t enough for agents

Until now, the question was simple: a human logs in with a username and password and is then allowed to do everything their account permits. This no longer works for agents. If I give my assistant program my password, it suddenly can do everything I can do. It could empty my account, even though I only wanted to order a pizza.

That’s why an agent needs its own identity with narrow limits. For example: the agent may shop with a merchant, but only up to 50 euros, only this week, and only with this one merchant. Experts speak of graduated permissions. Without something like this, every agent is a security risk that cannot be controlled.

It is also a matter of accountability. If an agent triggers a wrong order or sends thousands of requests to a server, it must be possible to trace who was responsible. Without identification, this is impossible. This is precisely why major payment providers, cloud providers, and standardization bodies are currently working intensively on this topic.

Credential, mandate, and protocol

Technically, agent identification usually consists of three parts. First, a credential for the agent itself, often a digital certificate or a cryptographic key. Cryptographic here means: the agent can mathematically prove something that no one else can forge. Second, a mandate proving that a particular person or company sent this agent out. Third, a list of permitted actions.

You can think of it like an intern with power of attorney. They have a company badge, a letter from their boss, and clear instructions about what they may purchase. The merchant checks all three things before handing anything over. If any one is missing, the transaction is rejected.

It is important to distinguish this from simple identifiers. Every access to a website already sends a kind of self-declaration today, the so-called user-agent string. It states, for instance, which browser is being used. However, this information can be freely invented, so it is not proof of anything. Genuine agent identification is verifiable because it relies on signatures that cannot simply be claimed.

Payment agents, bot blocking, and regulation

The issue becomes most visible in payments. Providers such as Visa, Mastercard, and several tech companies have introduced procedures that allow AI assistants to shop on their own. The agent is given its own digital identifier with a limit, instead of simply using the user’s card details. If the agent is misused, only this identifier gets blocked.

A second area concerns websites defending themselves against unwanted programs. Operators want to distinguish whether a helpful assistant is accessing on behalf of a genuine customer or whether someone is scraping content en masse. Only with reliable identification can one allow the former and slow down the latter. Some providers therefore already require agents to register and appear with a signature.

The term also appears in news about regulation. Regulatory authorities demand that automated decisions remain traceable. This includes a log that records which agent acted when and on whose behalf. Agent identification is therefore less a single technology than a basic prerequisite. Without it, autonomous programs can hardly be operated safely and in a legally sound manner.