- Telcos aren't prepared to deal with the risk of agentic AI
- AI agents are like water, they go where they want to go, said McKinsey Partner Duarte Begonha
- Human teams will be working with AI agents soon — it's time to get ready
ERICSSON OSS/BSS SUMMIT 2025, LONDON – AI agents "are like water, they go where they want to go," and telcos aren't prepared to deal with the risk of an unruly agent, warned McKinsey & Company Partner Duarte Begonha.
According to Begonha, telecom lacks risk mitigation processes unlike the banking industry, which has risk management down to a science. This lack of risk processes could put telcos at a disadvantage in the near future when agentic AI teams and human teams are working-side-by-side.
"Basically you need to ensure demand and confidentiality, to make sure that agents do not have access to do something that they're not supposed to do," said Begonha. "And yeah, sometimes agents can get excited and start doing stuff they're not supposed to do."
The 5 risks no one is talking about
Cross-agent task escalation: When one agent requests something from another agent, falsely escalating the task as coming from a higher up.
Sometimes agents can get excited and start doing stuff they're not supposed to do.Duarte Begonha, Partner, McKinsey & Company- Untraceable data leak: When two agents exchange data without oversight - and that exchange obscures data leaks.
- Synthetic identity risk: The forging or impersonating of an agent to bypass trust mechanisms.
- Chained vulnerabilities: A flaw by one agent cascades across tasks leading to flawed results.
- Data corruption propagation: Low-quality data is amplified by other agents which leads to distorted outcomes.
Steps to mitigate the risk
So what steps should telcos take to mitigate the risk? Begonha mapped out several things they must put into place and the panel following his presentation echoed his recommendations.
Telcos need to be able to set up guardrails around what agents to do, he said.. "You also need to be able to manage the daily changes that agents are doing and what agents are exposing."
They also need to understand what they need the agents to manage. Finding things that can be tracked easily is a good place to start — this avoids data corruption propagation.
When everything is well, everybody is happy, but if some of these things started to pop up, there will be, as always, “voices in the organization that they don't see the innovation. They see this with some fear.... It's a problem," said Begonha.