Podcast 149: AI in fraud is no longer offense vs defense, with Andre Isoni

For years, fintech has been defined by acceleration. Faster onboarding. Real-time payments. Automated compliance. The industry has consistently rewarded speed and efficiency.

This episode of the Fintech Garden Podcast reframes that trajectory. In a conversation with Andre Isoni, Chief AI Officer at AI Technologies, the focus shifts from acceleration to resilience. The question is no longer how fast systems can operate, but how well they can withstand adversarial pressure.

The discussion centers on a structural shift already underway.

AI Is Powering Both Sides of Fraud

Artificial intelligence is not just a defensive tool in financial services. It is actively used by attackers.

Fraud is no longer a one-sided problem where institutions deploy technology against static threats. It is a dynamic system where both sides evolve simultaneously. The same capabilities that enable fraud detection, identity verification, and risk scoring are being repurposed to bypass them.

Synthetic identities can now be generated at scale. Verification systems can be tested and probed automatically. Attack patterns can adapt in real time.

This creates a continuous feedback loop. Defenders improve systems. Attackers learn from them. The cycle repeats. The implication is clear. Fraud is no longer episodic. It is systemic.

Cybercrime Has Become Structured Innovation

A significant shift highlighted in the conversation is the professionalization of cybercrime.

Attackers are no longer limited to using existing tools. They are building their own infrastructure. Running internal research. Recruiting technical talent. Developing specialized solutions for specific targets.

This mirrors how legitimate technology companies operate.

The gap between attacker and defender is no longer defined by access to tools. It is defined by speed of iteration and absence of constraints. While financial institutions operate within regulatory frameworks, attackers do not.

This asymmetry creates a persistent disadvantage.

Security is no longer about preventing known threats. It is about anticipating engineered ones.

Smaller Models Are Changing the Risk Profile

Public discourse around AI often focuses on scale. Larger models. More data. Higher compute. The more relevant trend is the opposite.

Models are becoming smaller, more efficient, and easier to deploy. Capabilities that previously required large infrastructure can now run locally on consumer devices. This has direct implications for fraud.

Smaller models are harder to detect. Easier to distribute. Less dependent on centralized systems. As they shrink further, they become more invisible within existing monitoring frameworks.

Detection systems built around identifying large anomalies lose effectiveness in this environment.

Risk is not increasing because AI is getting bigger. It is increasing because AI is getting smaller.

Static Defenses Cannot Handle Dynamic Threats

Traditional fraud and security systems rely on predefined rules and historical data.

AI-driven threats do not.

Generative systems can modify their behavior continuously. Code can rewrite itself. Outputs can change while maintaining the same objective. This creates a fundamental mismatch between static defenses and adaptive attacks.

Detection becomes reactive by design. By the time a pattern is identified, it has already changed.

This forces a shift in approach. Security systems must become adaptive, flexible, and capable of responding in real time rather than relying on fixed logic. Predictability, once a strength in process design, becomes a vulnerability.

There Is No Silver Bullet in AI Security

The expectation that a single product can solve fraud is persistent and incorrect.

Every tool, whether AI-based detection, behavioral analytics, or encryption, represents only one layer of defense. None are sufficient in isolation.

Effective security is compositional. Multiple systems working together. Overlapping controls. Redundancy across layers.

This is not new in cybersecurity, but AI amplifies the necessity. As threats become more sophisticated, reliance on any single mechanism introduces risk.

The role of technology is not to eliminate fraud. It is to make it progressively harder.

Identity Verification Is Becoming Less Reliable

Digital onboarding has become standard across fintech. It is also becoming easier to exploit.

AI can generate realistic documents, replicate facial features, and simulate human behavior convincingly enough to pass many verification checks. This shifts identity from a verifiable attribute to a probabilistic signal.

The challenge is no longer validating data. It is establishing trust in the interaction itself.

This has implications across KYC, KYB, and account access. Systems designed under the assumption that identity inputs are stable are increasingly exposed. Verification must evolve beyond static checks.

Human Oversight Is Returning by Necessity

Full automation has been a core objective in fintech operations. AI complicates that objective.

While automation increases efficiency, it also introduces new failure modes. Systems can make incorrect decisions at scale. Synthetic inputs can bypass automated checks. Edge cases become harder to identify.

As a result, human oversight is re-emerging as a critical component.

Not as a replacement for AI, but as a complementary layer.

Review, validation, and exception handling require judgment that automated systems cannot fully replicate. This creates a hybrid model where AI handles volume and humans handle ambiguity.

Efficiency gains are preserved. Risk is managed.

Process Design Is Becoming a Security Lever

Beyond tools and models, the structure of processes itself is changing.

Traditional workflows in fraud detection and compliance are linear and consistent. This makes them efficient, but also predictable. Predictability allows attackers to learn patterns and design around them.

The shift is toward dynamic processes. Variable checkpoints. Non-linear verification paths. Adaptive escalation mechanisms.

By introducing variability, systems reduce the ability of attackers to anticipate outcomes. Process design becomes part of the defense strategy, not just an operational consideration.

Consistency optimizes operations. Variability protects them.

Behavioral Data Is Replacing Static Signals

As identity becomes easier to manipulate, behavioral data gains importance.

Transaction patterns, device usage, location history, and interaction timing provide a richer signal than static identifiers. These patterns are more difficult to replicate consistently, especially at scale.

Modern fraud detection increasingly relies on learning these behaviors and identifying deviations. This represents a shift in how trust is constructed. From who the user claims to be, to how they behave over time.

The effectiveness of this approach depends on data quality and the ability to continuously update models as behavior evolves.

AI Redistributes Cost Rather Than Eliminating It

One of the more pragmatic insights from the discussion is economic.

AI reduces the cost of execution. Tasks that required manual effort can be automated.

But the savings are not purely additive. They are reallocated.

As systems become more efficient, new costs emerge. Security. Monitoring. Governance. Compliance. Human oversight.

The net effect is not necessarily higher profitability, but increased operational complexity with improved scalability. AI enables growth. It also requires continuous investment to manage the risks it introduces.

Governance Becomes a First-Class Concern

The adoption of open-source AI models introduces additional responsibility.

Organizations using these models assume full accountability for compliance, data governance, and risk management. This is particularly relevant in regulated environments such as financial services.

Accessibility of technology does not equate to readiness for governance. As AI becomes easier to deploy, the burden of responsible use increases.

This creates a gap between what organizations can do and what they are prepared to manage.

Closing that gap is becoming a strategic priority. Speed defined the last phase of fintech innovation. Resilience may define the next.

As AI continues to evolve, the advantage will not come from adopting it fastest, but from integrating it in a way that balances efficiency, security, and trust.