Bryan Benson Warns: AI Trading Is Exposing Liquidity & Governance Failures in Crypto

Crypto platforms rarely collapse because of bad code. They usually break when users rush to move money and the infrastructure cannot keep up.

According to empirical research, liquidity spillovers across crypto markets increase significantly during times of market shocks and uncertainty. As a result, liquidity dries up, fiat rails slow, and governance decisions made during growth suddenly matter. That is the environment in which AI-driven trading and execution systems now operate.

As artificial intelligence (AI) accelerates execution and compresses reaction time, those structural limits surface faster.

Automation increases correlation across markets, amplifies small errors, and leaves less room for human judgment when conditions shift.

Few operators have seen those dynamics unfold from the inside.

Bryan Benson’s work scaling Binance across Latin America and later advising AI-driven trading and treasury automation systems placed him close to moments when infrastructure, governance, market assumptions and user behavior stopped holding at a global scale.

In this CCN interview, Dr. Lorena Nessi speaks with Bryan Benson, CEO of Aurum and former LATAM managing director at Binance, about how speed-first growth created long-term governance risk, what breaks first during market stress, why global user behavior defies simple models, and where AI-driven trading systems remain structurally fragile.

The Governance Gap: Why Speed-First Growth Becomes a Risk Multiplier

Binance’s expansion across Latin America took place in a region where crypto adoption surged unevenly, banking access varied sharply by country, and regulatory clarity developed at different speeds.

According to Chainalysis and World Bank data, several Latin American economies combined high crypto usage with limited access to traditional financial services, forcing exchanges to operate across fragmented payment rails and regulatory regimes simultaneously.

“The early growth playbook favored speed. Innovation was more agile than policy, and we built, learned, and adapted in real-time. That approach helped in a fast-moving industry when I worked as Binance’s LATAM Managing Director, yet scale turned it into a risk multiplier once operations expanded across Latin America’s very different markets, each with its own regulators, banking infrastructure, and consumer expectations.”

As operations expanded, governance did not automatically mature alongside user growth.

“What felt agile early on became harder to manage as the platform matured,” Benson said.

“Global systems demand global governance, and the operating model has to mature as fast as the user base does. Governance doesn’t ‘catch up’ on its own: once you’re global, the gap shows up early and everywhere at once,” he said.

Trust improved only when local leadership, native-language operations, and market-specific compliance were treated as core infrastructure rather than late-stage fixes.

Liquidations and Broken Rails: What Fails First When Markets Turn Volatile

Extreme market stress repeatedly exposes the same weak points. Liquidity fails before models adjust, Benson said.

“That level of stress exposes structural weak points,” he said. “Liquidity can evaporate faster than models and risk engines anticipate, and derivatives amplify the problem when margin and liquidation mechanisms feed volatility instead of absorbing it.”

As liquidation cascades build, execution quality deteriorates even for disciplined strategies.

Access rails fail at the worst possible moment. Fiat on- and off-ramps often become expensive, slow, or unavailable just as users need them most.

“Access rails also become fragile at the worst time,” Benson said. “Fiat on- and off-ramps can fail through poor localization or high costs, and user friction spikes precisely when seamless movement of funds matters most.”

According to Benson, fixes only arrived once exchanges adopted clearer controls, stronger localization, and governance models designed for stress rather than growth alone.

Global Users Do Not Behave as a Single Market

One of the most damaging assumptions at scale, Benson said, was treating global users as a single rational group.

“A global platform serves multiple user archetypes at the same time, and they behave differently for rational reasons,” he said.

He emphasizes that crypto functions as a practical financial tool.

“In many markets, digital assets function as practical tools for preserving value, managing inflation exposure, or keeping access to capital when traditional systems are unreliable,” Benson explained.

At the same time, speculative demand driven by leverage and volatility remains significant during peak cycles.

“A large cohort pursues risk and upside through leverage, volatility, and speculative trading,” he said.

Designing products as if those motivations were the same leads to mispriced risk and poor outcomes.

“Some people use crypto to get by; others chase leverage,” Benson said. “The product and the guardrails should treat those as different lanes.”

Where AI Introduces New Market Fragility

AI-driven trading is often framed as an efficiency upgrade. Benson said it also introduces new systemic risk.

“AI can compress reaction time across the market, and that speed can turn small signals into large, synchronized moves,” he said.

When multiple systems rely on similar data sources and strategies, crowding forms quickly and exits happen simultaneously.

“You also get fragility from what these systems depend on,” Benson said, pointing to exchange APIs, live price feeds, and cloud infrastructure.

Even minor disruptions can cascade through automated systems.

“If a WebSocket feed starts lagging, an API rate-limits, or a data stream glitches momentarily, you can end up firing off a chain of orders based on the wrong picture,” he said.

“AI removes the pause where a human might second-guess a trade,” Benson said.

Infrastructure vs. Automation: Why On-Chain Execution Can’t Be Fully “Clean”

Some elements of crypto markets remain structurally incompatible with clean automation.

“On-chain execution runs through public mempools, variable fees, and probabilistic confirmation,” Benson said, noting that strategies can lose execution to latency, reordering, or Maximal Extractable Value (MEV) dynamics.

Market data quality presents another constraint. Liquidity fragmentation, wash trading, inconsistent reporting, and maintenance windows all reduce reliability.

“AI can process noisy data, but noise still sets a ceiling on reliable execution,” Benson said.

In these environments, governance and safeguards matter as much as model performance.

Preventing AI Systems From Repeating Past Failures

“A model trained only on historical data can absorb yesterday’s market structure and overfit patterns that disappear during a regime shift,” Benson said.

That risk becomes more pronounced in crypto markets, where structural conditions can change quickly, invalidating learned patterns.

To address that, strong teams combine historical training with adversarial testing and post-training techniques.

Operational controls remain essential. Benson highlighted walk-forward testing, stress testing, drift monitoring, and hard risk limits.

“In production, drift monitoring, hard risk limits, and circuit breakers keep behavior bounded,” he said.

Even with these safeguards in place, the question shifts from how systems behave to who is accountable when they fail.

Responsibility, Disclosure, and User Impact

Benson stresses that accountability should align with control.

“Platform operators and model designers own accountability for what the system is built to do,” Benson said, including testing, risk limits, monitoring, and disclosures.

Users also bear responsibility for the risk settings and capital exposure they choose, while regulators define minimum standards for audits and disclosures.

That framework assumes all parties understand where control actually sits, an assumption that often breaks down at the retail level.

Retail users, however, often lack visibility into AI-driven execution.

“Consumer-facing systems have an ethical responsibility to create legibility around the ways AI is influencing the execution of trades,” Benson said.

Disclosures should clearly explain where AI has control, when systems reduce activity, and how conflicts are handled. Without that transparency, users cannot meaningfully assess risk or give informed consent, turning automation into an ethical issue rather than a technical one.

This is where the conversation extends beyond engineering and into broader debates among AI experts, regulators, and social scientists about power, agency, and accountability in automated systems.

As AI increasingly mediates financial decision-making, questions of ethics, incentives, and oversight become harder to separate from market design itself.

Questions the Industry Avoids

Benson said incentive alignment remains one of the most uncomfortable issues in AI-native finance.

“Many products benefit from higher turnover, wider spreads, or user overconfidence,” he said.

Auditability, data rights, and systemic concentration also remain unresolved.

“When many platforms depend on the same cloud vendors, the same market data feeds, and similar models, a single failure mode can spread quickly,” Benson said.

For AI-native finance to mature, shared norms around transparency, accountability, and stress testing must evolve alongside technology.