KPMG and Cequence Set Runtime Controls as the Agentic Governance Baseline
Source
Delegating Outcomes, Not Authority: Agentic AI Runtime Security, Governance, and ControlCequence Security and KPMG
What happened
Cequence Security and KPMG in Singapore released Delegating Outcomes, Not Authority: Agentic AI Runtime Security, Governance, and Control in November 2026. The joint report presents a framework for governing AI agents during execution, not only at the point of access provisioning. The report draws on Cequence field data showing 281,627 AI agent visits across five industries in the week of October 15-21, 2026. Roughly one in seven agent visits took a consequential action such as booking, purchasing, signing in, or managing an account. Separately, at one telecommunications carrier, unidentified agent traffic grew from about 100 requests per day to as many as 32,000 daily. This growth occurred within three weeks of a major consumer agent launch. Those agents passed multi-factor authentication by signing in as customers. The report also documented a credential stuffing campaign at an online marketplace in which fraudulent traffic claimed to be a well-known consumer AI agent. The two firms plan to present findings to business, technology, and policy leaders at KPMG's Singapore office on November 19, 2026.
Why it matters
- ·Traditional access controls confirm who an agent is and what systems it can reach. They cannot determine whether a specific action at a specific moment is within the agent's authorized scope. As 40% of Enterprises Let AI Act Without Human Approval, Deloitte Finds has shown, the gap between granted permissions and appropriate actions is a live compliance exposure, not a future risk.
- ·The credential stuffing campaign in the report shows agent impersonation is already occurring at scale. Fraudulent traffic successfully mimicked the identity signature of a trusted consumer agent to reach sign-in pages. Compliance teams whose vendor risk programs rely on agent identity alone to establish trust have a control gap that the report's field data makes concrete.
- ·Runtime evidence is becoming what regulators and auditors ask for first. The report's framework aligns with the direction flagged in 1,000+ AI Rules, One Audit Gap: Runtime Evidence Is What Regulators Want. Organizations that cannot produce logs of agent action decisions, not just access logs, may struggle to demonstrate control adequacy under existing and emerging frameworks.
Stay ahead of stories like this
Get every Singapore AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- Containment failures across AI agents, vendor privacy rollbacks, and deepfake fraud show that unverified trust in AI systems is now a measurable financial liability.12 Nov
- State-linked attackers and criminal operators are actively weaponizing AI tools as entry points into enterprise systems, turning internal platforms and open-source frameworks into live attack vectors.5 Nov
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Ask your security and technology teams whether your AI agent governance program evaluates each agent action before it executes, or only checks permissions at login and deployment time.
- ☐Request a list of all AI agents currently operating in production that authenticate as or on behalf of customers or employees, and confirm whether those sessions are distinguishable in your access logs from human sessions.
- ☐Review your existing audit trail requirements for AI agents to confirm they capture what action was taken, what decision was made about that action, and by what control, not just that an agent was active.
- ☐Direct your vendor risk team to assess whether any third-party AI agents your organization relies on could be impersonated by external actors using the same identity signals, based on the credential stuffing pattern documented in this report.
- ☐Check whether your incident response procedures address agent impersonation as a distinct threat scenario, separate from account takeover involving human credentials.
What to watch next
The KPMG and Cequence Singapore launch event on November 19, 2026, may produce additional practitioner guidance that compliance teams should monitor. Singapore's regulatory environment for agentic AI is active. The Singapore Model AI Governance Framework for Agentic AI already addresses agent-specific controls voluntarily. The Five Eyes Guidance on the Careful Adoption of Agentic AI Services has set a security baseline for government-adjacent deployments. As regulators globally shift from policy guidance to enforcement, organizations should expect runtime evidence requirements to become audit standards sooner than governance programs anticipate.