PRNewswire
Singapore, November 11: Cequence Security, the chief in software, API, and agentic AI safety, at present printed Agentic AI Runtime Security, Governance, and Control, a report on governing AI brokers as they act. This report has been developed with enter from KPMG in Singapore, inspecting how organisations can govern and safe Agentic AI as brokers transfer from producing content material to autonomously taking consequential actions. The findings of the report shall be offered to senior enterprise, expertise, and coverage leaders at a joint occasion hosted at KPMG’s workplace in Singapore on 19 November. Read the chief temporary or obtain the complete report.
AI is shifting from answering inquiries to taking motion, with brokers now issuing refunds, updating information, and transferring cash. The report units out how one can govern these actions as they occur, for brokers deployed by organizations.
What Cequence’s evaluation and information present
– Agents already act for patrons: In one week of October 2026, Cequence’s CQ Prime analysis group tracked greater than 1 / 4 million AI agent visits throughout 5 industries. About one in seven booked, purchased, signed in or managed an account for the client.
– Many by no means say they’re brokers: At a Tier 1 telecommunications provider, visitors from brokers that did not establish themselves as brokers grew from about 100 requests a day to as many as 32,000 inside three weeks of a serious client AI agent’s launch. They signed in because the buyer and handed multi-factor authentication.
– A trusted title proves nothing: At one on-line market, extra visits claimed to be Google-Agent than all the real agent visits Cequence tracked that week. Every go to went straight to the sign-in web page and failed, an indication of attackers testing stolen passwords in bulk below a official agent’s title. Google was not concerned.
What Cequence recommends
– Give every agent a particular job, not open-ended authority: Each agent ought to work from a plain-language job description became onerous limits, so it will possibly’t determine by itself to do extra.
– Check each motion in actual time: A separate management exterior the agent ought to approve, escalate, or block every motion in actual time, since harmless-looking steps can add as much as one thing the agent is not allowed to do. When it blocks an motion, it ought to level the agent to the allowed route so the work nonetheless will get executed.
– Measure success by work accomplished, not actions blocked: Good governance and safety let brokers get extra official work executed with out the organisation shedding management.In a managed examine utilizing a Cequence analysis prototype, brokers with solely logins and stuck permissions tried to behave exterior their job in 44% of runs. When every agent was given an outlined job and each motion was checked towards it, these makes an attempt fell to 1%; process completion rose from 76% to 95%, and brokers used about 35% fewer tokens per run.
‘Organisations throughout Asia-Pacific are transferring from experimenting with AI to deploying them to realize actual enterprise outcomes. Traditional controls corresponding to identification verification and stuck entry permissions are essential, however they is probably not enough when an agent can perform a sequence of actions on behalf of a enterprise or buyer. Organisations want governance that considers the agent’s position, the context of every motion, and the cumulative threat created throughout a whole process. KPMG’s contribution to this report brings collectively our expertise in governance, threat, regulation and cyber safety to assist organisations put acceptable oversight round AI brokers whereas enabling official work to proceed,’ mentioned Gerry Chng, Partner, Head of Cyber, KPMG in Singapore.
‘An approval at the first step should not carry into step 5,’ mentioned Shreyans Mehta, CTO and co-founder of Cequence Security, who co-authored the report. ‘Runtime governance means checking what an agent does at each step. When the reply isn’t any, it means telling the agent why, so it will get again on monitor as an alternative of in search of a means round. It has to work the identical means for an agent your buyer sends as for one you constructed your self.’
Expanding in Asia-Pacific
Cequence’s presence is rising in Asia-Pacific, the place it has established its regional headquarters in Singapore. The firm has a regional group, together with senior management in Australia, with plans to construct native groups within the Philippines, Singapore, and South Korea over the approaching months. In the area, its prospects embody NTT DOCOMO in Japan and Philippine Airlines.
‘AI brokers are already performing for patrons throughout the area, and boards need to know they’ll belief them,’ mentioned Ameya Talwalkar, CEO and co-founder of Cequence Security. ‘KPMG has a deep understanding of how boards and regulators take into consideration threat, and Cequence sees what brokers truly do on our prospects’ functions and APIs day by day. Together, we are able to present organisations how one can give brokers actual work with out shedding management. Singapore is setting the tempo on governing AI brokers, which makes it the pure residence for our Asia-Pacific enterprise.’
About Cequence Security
Cequence protects the functions and information that energy the agentic enterprise. More than a decade of bot protection and API safety expertise has established Cequence because the chief of secure and safe agentic AI adoption. The Cequence platform delivers deep perception into person, entity, and agent habits, enabling organizations to safe, govern, and management agentic AI workflows whereas defending towards unhealthy actors and rogue brokers. Cequence delivers worth in minutes slightly than days or even weeks with a extremely scalable, no-code method. Trusted by the most important and most demanding non-public and public sector organizations, Cequence protects greater than 10 billion each day API interactions and 4 billion person accounts. To study extra, go to: https://www.cequence.ai.
(ADVERTORIAL DISCLAIMER: The above press launch has been offered by PRNewswire. ANI is not going to be accountable in any means for the content material of the identical.)