Written by a human
Is the secret to AI compliance treating it like a human?
As regulators from the SEC to the FCA sharpen their focus on AI and accountability, the firms getting AI right are the ones treating it like a workforce, not a fix.
Written by a human
In brief:
- Firms have rushed to roll out AI, but for many, transformation is hitting roadblocks with governance, trust, and explainability
- Successful implementation hinges on a leadership-led strategy that is outcomes first, technology second
- Organizations expecting human-like outcomes from AI need to consider if their expectations around accountability and compliance must also meet a “human standard”
Our relationship with AI is beginning to feel like it is well past the “honeymoon period.” The rose-tinted glasses of the opportunities AI present for financial services firms are still in place for many, but early adopters are now facing the challenges of making their “AI relationships” work day-to-day. From proving return on investment to establishing effective governance frameworks, opportunity is now meeting reality.
Sahar Kayhani, Global Relay’s Chief Product Officer, recently sat down with KPMG at their AI Made Real Summit to talk about how firms can keep the “shiny new Ferrari” that is AI moving towards their destination. Everyone wants their “dream car,” but few people have thought about who is driving, where the road leads, or what happens if it breaks down.
Sahar explored how the following conditions must be met for firms to truly optimize their AI:
- Staying up to speed with shifting regulatory standards as AI evolves
- Ensuring that strategy is leadership-driven
- Using explainable chain-of-thought reasoning
- Keeping a human in the loop with Agentic AI use
“Outcomes first, technology second”
Concerning leadership-driven AI strategy, Kayhani emphasized that it must move in two directions at once, stating that “leadership needs a clear mandate and strategy” to avoid enablement stalling before it truly begins.
However, this strategy must also filter down into how employees work day-to-day, or risk remaining unactioned. Internally, this looks like investing in faster and more accessible training across the entire business, so teams understand what effective AI utilization looks like in practice, and what the intended goal is. AI holds huge potential to deliver; firms need to maintain oversight of the intended outcomes of AI use so they can benchmark its true effectiveness.
Chain of thought, chain of trust
Chain-of-thought reasoning is more than a technical feature, and Kayhani explored how it works to enhance the outputs of large language models (LLMs), particularly for tasks involving multi-step reasoning. Chain of thought breaks down complex problems into manageable, logical intermediate steps that sequentially lead to a conclusive answer, meaning that the user has full oversight of each step a model has taken to reach a decision.
When a system can evidence why it flagged a risk or an alert against a firm’s risk taxonomy, that transparency serves two important purposes. First, it gives a clear an explainable audit trail that firms can rely on should regulators or legal entities request it. Second, it doubles as built-in ongoing training for compliance teams who are looking to become more knowledgeable in the AI field and allows firms to optimize their AI models by reviewing these processes and using this knowledge to streamline prompts and inputs.
Holding AI to human standards
Many expect AI models to be able to perform as well as, if not better than, humans in similar roles. But in order to progress AI efficiently and compliantly, Sahar urges the industry to “consider AI agents as digital workers in a similar fashion that human employees are.”
This reframing is important as, when AI moves from assistant to actor, questions around trust and accountability materialise quickly. Every time AI “touches” data or makes a decision, there must be an audit trail. Every action gets logged and recorded. This is particularly vital where Agentic AI is deployed, with every action that an agent takes subject to the same surveillance and risk identification process as a human operative. Simply put, Agentic AI will require firms to move towards “agentic governance.”
The regulatory bar may keep moving
Regulators are yet to make any official moves towards specific AI regulation, with 48% of firms seeing this hesitancy as something of a barrier to AI adoption. However, the pace of transformation AI is driving across financial services will mean regulators will move eventually – and firms must put in groundwork now to get ahead of whatever these movements might look like.
The Securities and Exchange Commission (SEC) has spent the past couple of years soft launching an enforcement positioning around “AI washing”, where firms are overstating the capabilities of their AI products. Similarly, the Federal Trading Commission (FTC) has also setup a dedicated unit to combat this same issue. =
The Financial Conduct Authority (FCA), has no specific AI rulebook, however, is an advocate for safe adoption of AI in UK financial markets to encourage innovation, and has recently taken a step towards potential AI-specific legislation.
Concerningly, joint FCA research with the Bank of England found that a small proportion of AI use cases in UK financial services currently run without a human in the loop for individual decisions – something we may well see specific regulatory messaging or expectations around emerge.
Building AI means building trust
Whatever applications firms are leveraging AI for, in order to build effective AI that is performant and integrated compliantly across the organization, it all comes down to trust:
- Trust that a leadership-driven strategy, effectively communicated across the business, will ensure that AI use is targeted at delivering outcomes
- Clear chain-of-thought reasoning allows firms to build trusted, transparent models, as they have a consistent and clear log of why decisions were reached
- Developing “agentic governance” for Agentic AI that holds AI to the same standards of oversight and accountability as a human will help build trust of model efficacy across your organization, and regulators to trust you are “compliance first” on AI
- Keeping up to speed with regulatory changes around AI will allow firms to continually audit and assess their AI use to ensure that it is in step with – if not in front of – AI regulation as it develops
To integrate a surveillance solution that allows for proactive AI responses with thorough and comprehensive oversight, click here.
Tags
Global Relay Connectors
Archive data from any channel. Explore our range of connectors to collect, store, and access your communication data for complete compliance.