Cybersecurity
Cequence and KPMG Turn Attention to the Growing Security Risks of AI Agents

New research from Cequence Security finds AI agents signing into accounts, booking services and making purchases on behalf of customers, highlighting the growing difficulty of controlling automated activity across business platforms.
[For more news, click here]
AI agents are beginning to handle tasks that have traditionally required direct human involvement, including signing into customer accounts, booking services and making purchases. As businesses introduce these systems into customer-facing services and daily operations, their ability to act on behalf of users is creating new questions around access, accountability and security.
Research by Cequence Security shows how widespread some of this activity has already become. Its CQ Prime research team tracked more than a quarter of a million AI agent visits across five industries during one week in October 2026. Around one in seven involved agents booking, buying, signing in or managing customer accounts on behalf of users. Some were also operating without identifying themselves as AI agents.
At a Tier 1 telecommunications carrier, traffic from agents that did not identify themselves as automated systems increased from approximately 100 requests per day to as many as 32,000 within three weeks of a major consumer AI agent's launch. According to Cequence, these agents signed in as customers and passed multi-factor authentication. An online marketplace presented another concern, with visits claiming to be Google-Agent outnumbering all genuine agent visits tracked during the same week. Every one went directly to the sign-in page and failed. Cequence believes the pattern indicated attackers testing stolen passwords in bulk under a recognized agent identity.
These findings form the basis of a report Cequence developed with input from KPMG in Singapore. Titled Agentic AI Runtime Security, Governance, and Control, it examines how organizations can govern agents as they move beyond generating content and begin carrying out actions that affect customers and business operations.
Controlling what AI agents can do
Cequence and KPMG recommend assigning each agent a clearly defined job and translating its responsibilities into enforceable limits. They also propose a separate control mechanism that assesses actions in real time and can approve, escalate or block requests. When an action is rejected, the system should explain why and direct the agent toward an authorized alternative, allowing legitimate work to continue. The report recommends measuring success by the work agents complete rather than the number of actions blocked.
A controlled study using a Cequence research prototype illustrates the potential impact. Agents operating with only login credentials and fixed permissions attempted to act outside their assigned jobs in 44% of runs. When given defined responsibilities and subjected to checks against those responsibilities at every step, the figure fell to 1%. Task completion increased from 76% to 95%, while agents used approximately 35% fewer tokens per run.
"An approval at step one shouldn’t carry into step five," said Shreyans Mehta, CTO and co-founder of Cequence Security, who co-authored the report. "Runtime governance means checking what an agent does at every step. When the answer is no, it means telling the agent why, so it gets back on track instead of looking for a way around. It has to work the same way for an agent your customer sends as for one you built yourself."
The recommendations cover both agents developed internally and those introduced by customers, which may access business applications and APIs using customer credentials. This distinction is particularly relevant for organizations that do not have direct control over every automated system interacting with their platforms.
Cequence expands in Asia-Pacific
The report also comes as Cequence strengthens its regional presence, with Singapore serving as its Asia-Pacific headquarters. The company has senior leadership in Australia and plans to establish local teams in the Philippines, Singapore and South Korea over the coming months.
"AI agents are already acting for customers across the region, and boards want to know they can trust them," said Ameya Talwalkar, CEO and co-founder of Cequence Security. "KPMG has a deep understanding of how boards and regulators think about risk, and Cequence sees what agents actually do on our customers’ applications and APIs every day. Together, we can show organizations how to give agents real work without losing control. Singapore is setting the pace on governing AI agents, which makes it the natural home for our Asia-Pacific business."
The findings will be presented to senior business, technology and policy leaders at a joint Cequence and KPMG event at KPMG's Singapore office on November 19. The executive brief and full report are available through Cequence Security.
Related Articles
Cequence Security Launches Agent Personas to Automate AI Agent Governance Across the Enterprise
Delinea Takes AI Agent Security Beyond Access
AI Agent Governance Is Falling Behind Enterprise Adoption, Optro Research Finds





