Archive · 12 Oct 2026
Cognizant Cybersecurity Chief Warns of Rogue AI Agent Risks
AI agents are computer programs that can make decisions and take actions with less help from people.
Vishal Salvi of Cognizant says this can create new security risks.
An agent might be tricked or given too much access, and then do something it should not.
Companies should decide what each agent is allowed to see and do.
They should also watch agents while they work and be able to stop them.
Agents can do simple, repeatable security tasks, but people should approve important actions.
Salvi says companies need more than ordinary login and access controls.
They should also keep track of how agents behave and what information shapes their decisions.
Cognizant says it helps businesses build and operate AI systems with these safeguards.
Vishal Salvi says autonomous AI agents can become a security threat if manipulated, compromised or given excessive permissions.
He recommends setting access and action limits during design, then monitoring agents continuously while they operate.
Salvi says AI can handle bounded, repetitive and reversible cybersecurity tasks, while people retain oversight of consequential actions.
He says identity controls and Zero Trust are necessary but must be supplemented by behavioural monitoring, context integrity and AI governance.
Cognizant says it helps clients secure AI systems through development controls, identity protections, monitoring, response and responsible-AI governance.
- Who
- Vishal Salvi, global head of Cybersecurity Services at Cognizant, discussed AI-agent security in an interview with Sudhir Chowdhary.
- What
- Salvi described security risks from autonomous AI agents and safeguards enterprises can use.
- When
- The interview's publication date is not stated.
- Where
- The interview's location is not stated.
- Why
- Autonomous agents can reason, access systems and act with limited human intervention, creating risks if manipulated, compromised or given excessive permissions.
More Agent Autonomy
Human Oversight and Control
How much autonomy to allow
More Agent Autonomy
AI agents can automate repetitive, bounded and reversible cybersecurity work, providing speed and scale.
Human Oversight and Control
Human oversight should remain mandatory for actions with significant business, customer, financial, safety or regulatory consequences.
Security controls for agents
More Agent Autonomy
IAM and Zero Trust remain essential foundations for securing agent access.
Human Oversight and Control
Those controls alone are insufficient; Salvi calls for continuous behavioural monitoring, context integrity, AI-specific governance and evidence of agent actions.
- Interviewee
- Vishal Salvi, global head of Cybersecurity Services, Cognizant
- Interviewer
- Sudhir Chowdhary
- Key concern
- Agents may be manipulated, compromised or granted excessive permissions.
- Recommended approach
- Set controls at design time and enforce them continuously at runtime.
- Human oversight
- Salvi says people should approve actions with material business, customer, financial, safety or regulatory consequences.
- Existing controls
- Identity and access management (IAM) and Zero Trust are essential foundations, but are not sufficient alone.
- Cognizant’s stated support
- Secure AI development, AI detection and response, identity controls, context-integrity protections, monitoring and responsible-AI governance.