Scroll
Before and After · Differences
The agent uses the credentials of the person responsible.
Undefined limits
Indistinguishable Acts of the Person
No independent proof
The agent has his own identity and limits
Signed authorisation (QTSP)
Checked before each action
Sealed and verifiable receipt
*Applies to actions mediated by ds Agent Guard|Agent.
DigitalSign Qualified Service
That is, a document signed by DigitalSign that says what the agent can and cannot do.
Then it checks each request and decides: allow or refuse the action.
Finally, each decision has a qualified temporal seal and is therefore auditable by third parties.
The AI can be manipulated. However, what she can do does not change. And every decision is sealed and verified.
Each order passes first by Agent Guard and then leaves a sealed receipt. This way, there are four pieces and one check per request.
4 Steps · 2 Phases
IT team, legal, risk and business.
Responsible nominee and DigitalSign QEAA.
Agent Guard: allows or refuses each action.
Sealed receipt and time stamp, audited.
So who authorized the limits, when and with what scope is proved. Consequently, this helps to demonstrate diligence and limits exposure to civil liability, albeit without eliminating it.
Anything that is not allowed is refused by default.
QEAA signed by DigitalSign under eIDAS 2.0.
Validity status verified in real time.
Receipts with qualified temporal seal, verifiable.
Limits defined by multidisciplinary team and authorised by administration.
Complements IAM, API gateways and sandboxes. Python/Node SDK.
Stage 1
For example, a customer asks about reading the counter.
Allowed
Scenario 2
On the other hand, someone asks for the NIF and the status of another client's account.
Denied
Scenario 3
In a complaint, the agent obeys and tries to change data.
Denied
Scenario 4
Likewise, two agents try to send out data.
Denied
Note, however, that this is a demonstration environment: fictitious company, easy to manipulate internal AI model, QEAA in pre-production.
DS AgentID complements and does not replace sandboxes, network controls, IAM and API gateways. Therefore, for full protection, we recommend agents without their own credentials, denied network output by default, closed failure and authorisations with the minimum of necessary privileges.
The QEAA are a trusted qualified service introduced by eIDAS 2.0. In particular, in accordance with Article 45 and Annex V to the Regulation, they contain the information necessary to identify the issuer, the holder, attributes and period of validity. Thus, in DS AgentID, the attribute attested is the agent's authorization.
The DS AgentID may support organisations in fulfilling governance obligations. For example, Directive NIS2 (Article 20) requires administrative bodies to approve cybersecurity risk management measures and supervise their implementation. Similarly, the IA Regulation (Article 26) requires, in high-risk systems, human supervision by persons with competence, training and authority, and the preservation of the records generated.
The application to each organisation shall be assessed by the respective legal department.
DigitalSign accreditation as QTSP and for the emission of QEAA can be confirmed in EU Trusted List. The qualified service is the issue of certificates and QEAA, stamps and time stamps; Agent Guard and SDKs are the software that applies that trust in the organization's systems.
Not in a perceptible way. For example, in our laboratory, encryption and politics take less than 1 ms and a full order takes between 26 and 36 ms (median and 95th percentile). In comparison, a step of the AI model takes between 1 and 10 seconds. Note, however, that the measurement was made in a prototype with 500 calls and test credentials.
Yeah. Even in simple tasks, without formal authorization the agent acts with the credentials of the person responsible. Consequently, their acts are confused with those of a person and the scope has never been defined or authorized by anyone. On the contrary, with DS AgentID the agent has its own identity and limits and, in addition, each action is proven.
As little as possible. In practice, DS AgentID integrates through an SDK (Python and Node) or Skill or MCP for agent frameworks. In addition, Agent Guard can be installed in a central gateway, a gateway by service or within MCP itself.
Nope. On the contrary, it complements them. These layers help to contain the agent in your environment. In turn, DS AgentID adds a third party authorisation and verifiable proof.
If Agent Guard becomes unavailable, the agent will stop acting. That is, without verification, the actions are refused instead of going without control (closed failure). Besides, a compromised agent can't erase the evidence. This is because each decision remains on a sealed receipt, with a qualified time stamp, kept outside the agent, and therefore any subsequent change would be detected.
Nope. However, it helps demonstrate diligence and reduces the damage that an agent can do through actions mediated by Agent Guard, delimiting exposure without eliminating it. In any case, the legal assessment of each case is the responsibility of the organisation.
First, the issue is licensed per agent per year. Then Agent Guard works by subscription, with real-time revocation status. Finally, time stamps and audit are provided by volume as a service.
The AI is powerful. Just limit what she can do. Prove what you did. Talk to us about how to set the first limits of an agent, issue theirs (QEAA) and get Agent Guard to check each action, and see DS AgentID working in a demo.
DS AgentID integrates into the same digital identity ecosystem as DigitalSign.
It applies to actions mediated by DS Agent AgentID. General information, which does not constitute legal advice.
