
Data, access and agents protected under the company's rules
An email from a supplier can contain instructions written specifically to trick an agent into forwarding a confidential document. With Security we give each agent an identity and only the permissions its task requires. We work alongside the people responsible for IT security in the organisation and, with them, assess the level of risk and the measures to adopt, process by process.
Security · What it contains
- Identity and permissions
- One identity per agent, with only the permissions of its mandate, inherited from the source systems
- Model access
- A single point for requests to models: which are allowed, with which data and spending limits
- Data protection
- Masking of confidential data before it is sent to a model, encryption, separate environments
- Agent defence
- Checks on what the agent reads and produces; the permitted tools are a closed list
- Credentials
- Keys and passwords in a protected vault: the agent uses them without seeing them
To security, an agent is a new kind of user
Like a user, the agent reads documents and acts on behalf of a person or a department, but it works without breaks and can be deceived by what it reads. That is why we check any action that a piece of content suggests to it against its mandate.
What it contains
Agent identities go into the system that manages users, and events reach the monitoring already in use, so those who run security do not have to learn one more console.
01
Identity and permissions
One identity per agent, with only the permissions of its mandate, inherited from the source systems
02
Model access
A single point for requests to models: which are allowed, with which data and spending limits
03
Data protection
Masking of confidential data before it is sent to a model, encryption, separate environments
04
Agent defence
Checks on what the agent reads and produces; the permitted tools are a closed list
05
Credentials
Keys and passwords in a protected vault: the agent uses them without seeing them
06
Monitoring and shutdown
Events go to the company's monitoring tools; an anomalous agent is suspended at once
Who sets the rules and who enforces them
| Governance | Security | |
|---|---|---|
| Who it speaks to | GovernanceRisk, compliance, internal audit | SecurityIT security, information systems |
| An example | GovernanceThe supplier invoice agent may read orders, not payslips | SecurityThe agent is unable to read them, and the attempt is reported |
Illustrative example
What is decided beforehand
We examine the options with those accountable for security and data protection before development starts, because changing them once the project is under way costs far more. There is also a limit that holds for any configuration: the module's controls apply the permissions it receives, and none of them replaces the periodic review of those permissions by those who govern the company's systems.
Frequently asked questions
Does our data leave our systems?
It depends on the environment you choose: your own cloud, a European cloud or an isolated environment. In no case do we use it to train third-party models.
Does it replace our security tools?
No. It extends to agents the controls you already apply to people and applications, and sends events to the tools you use today.
Does it also apply to agents from other suppliers?
Yes, provided they have a registered identity and go through the single point of access to models.
Examine the security model with those accountable for it
Bring to the table the people who manage identity, monitoring and data protection, and we will look together at how to apply your rules to agents.