Aveni Proves Real-time Agent Assurance in FCA’s Inaugural Supercharged Sandbox as Firms Race to Deploy AI Agents

Aveni, the AI fintech, has completed a pilot of its Agent Assure solution in the Financial Conduct Authority’s inaugural Supercharged Sandbox. Aveni has demonstrated how end-to-end assurance — spanning pre-deployment stress testing and post-production monitoring — can unlock safe deployment of agentic AI in regulated environments.

The milestone comes as new industry data reveals a growing trust gap holding back the adoption of autonomous AI agents in financial services. While appetite for autonomous AI agents is near universal, real-world deployment remains limited. Research shows that 99% of companies plan to put AI agents into production(1), yet only 11% have actually done so. At the same time, just 2% of companies report having adequate AI guardrails in place (2), while 95% have already experienced at least one AI-related incident — 77% of which resulted in financial losses (3). Together, these figures underline a critical challenge for the industry: ambition is accelerating faster than assurance.

The introduction of autonomous AI agents has the potential to fundamentally transform consumer financial services through highly personalised, always-on guidance. However, greater personalisation also introduces greater risk. As agent deployments scale, firms face a surge in interactions requiring oversight, while traditional assurance models — reliant on sampling, lagging reviews, and human-intensive processes — struggle to protect customer outcomes at the point they matter most.

Against this backdrop, Aveni’s work in the Supercharged Sandbox focused on evidence-led assurance — using simulated, real-world interactions to test AI agent behaviour and demonstrate compliance with safe conduct standards prior to deployment, alongside continuous production monitoring to provide scalable oversight once live.

Joseph Twigg, CEO of Aveni, said:“We need to get ready for the next phase of assurance: agentic assurance. This will be a complete paradigm shift for the industry. The industry is rushing towards autonomous AI agents, but the uncomfortable truth is that most firms cannot yet prove those agents are safe. 

Twigg continued: “Our work in the Supercharged Sandbox with Agent Assure creates a new approach, delivered live and using small language models to act as independent assessors of primary models. It has been designed around the principle of unified assurance. From a regulatory perspective, it does not matter whether a customer is speaking to a human adviser or an AI agent — what matters is that the outcome is fair, safe, and compliant. As firms increasingly blend human and agent-led interactions, consistency of assurance across both becomes essential.”

During the Sandbox, Aveni piloted the core components of its Agent Assure solution using an agentic framework: a first-line protection guidance agent grounded in real client use cases and customer pain points, alongside a two-tier assurance capability. 

Following the Sandbox, Aveni plans to extend these methods into live environments, scaling oversight across real decision streams using enterprise-grade infrastructure. The next phase includes completing a test harness to enable pre-deployment validation: firms can now upload their AI agents to Agent Assure to be stress tested before they go live. 

These simulated interactions are assessed to identify potential failure modes — such as weak vulnerability handling, inconsistent outcomes, or unsupported claims — and are packaged into independent, evidence-based risk reports. This gives both agent-building teams and second-line risk and compliance functions a shared, objective view of potential customer harm, enabling issues to be remediated and retested before deployment.

In parallel, Aveni will further fine-tune its suite of UK finance-specific language models – FinLLM – to support real-time guardrails that can take proportionate action inserting required disclosures, steering conversations to safer flows, or blocking high-risk outputs. This will ensure issues are prevented, not just observed. 

Every implementation faces trade-offs between speed, safety, and customer experience and Aveni is working with partners to find the right balance for each use case, ensuring controls are proportionate rather than one-size-fits-all. 

An application to the FCA’s AI Live Testing programme will be made for Agent Assure before wider deployment to ensure it is regulatory-ready and operationally robust.