Most AI agents operate inside third-party platforms where external gateways cannot inspect activity.
Examining how agentic workflows rely on host permissions, runtime guardrails, and third-party integrations across enterprise environments.
A recent example shows how an employee asked a digital assistant to send a mass email. When the process was interrupted due to sending limits, the assistant, because it also had access to the employee’s personal email, suggested sending the messages from there. The employee approved the action, resulting in messages being sent from a personal account. The solution to this problem was provided by a simple rule that blocked attempts to contact over fifty external addresses and explained the tool’s actions.
Where does the danger lie?
Large platforms, such as customer management tools, communication apps, and document services, are now integrating digital assistants. There, these programs use the features, data, and permissions of the respective platform. Security systems that only control traffic at central network points are not enough, because assistants operate internally and the data does not pass through the points monitored by the company.
Additionally, the tools that manage AI traffic themselves can be targeted. In March 2026, malicious packages were detected in LiteLLM, an open-source tool that directs AI traffic. The issue remained active for approximately forty minutes. According to CloudSEK estimates, more than 2.500 organizations were exposed during its duration.
How can companies be protected?
To be truly protected, companies need to know exactly what is going on in their systems. A suggested ninety-day security process includes the following steps:
- Weeks 1 to 2: Identify all active digital assistants.
- Weeks 3 to 6: Assign responsibilities and control the settings for each tool.
- Weeks 7 to 12: Create security rules depending on the severity and type of data.
Some companies, like Reco, offer turnkey detection services that use special read-only digital keys to deliver a comprehensive security picture and remediation plan within a few business days.
↳
READ ALSO
HACKS.GR / RELATED
Cyber Attacks & Data Breaches
Three members of the VasyGrek hacker group arrested in Russia
↗
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
Top comments (0)