SKN | Researchers Warn AI Agents Must Be Treated as Untrusted Systems as Security Risks Escalate Across Automated Infrastructure

Key Takeaways

• Security researchers are urging developers to treat AI agents as inherently untrusted systems due to rising risks of manipulation and autonomous exploitation.
• The warning arrives as AI-driven automation expands across crypto trading, DeFi execution, and blockchain infrastructure tooling.
• The findings raise new concerns about systemic vulnerabilities as AI agents increasingly interact with financial protocols handling billions in daily volume.

Researchers in cybersecurity and artificial intelligence have issued a warning that AI agents should be treated as untrusted systems by default, highlighting growing concerns about their susceptibility to manipulation, prompt injection attacks, and unintended autonomous behavior. The report comes as AI integration accelerates across digital asset markets, where algorithmic trading systems already account for a significant share of daily crypto transaction volumes, often exceeding tens of billions of dollars globally across spot and derivatives markets.

The findings arrive at a time when both centralized exchanges and decentralized protocols are increasingly deploying AI-driven systems for trading optimization, customer interaction, and automated smart contract execution. As crypto markets continue to evolve alongside artificial intelligence infrastructure, the overlap between financial automation and autonomous decision-making systems is becoming more pronounced.

Rising Exposure Across Crypto and Blockchain Systems

According to the researchers, AI agents operating in connected environments can be exposed to malicious inputs that alter their behavior in unpredictable ways. In crypto markets, this risk is amplified by the composability of decentralized finance systems, where smart contracts interact continuously with external data feeds, automated bots, and off-chain AI models.

Daily trading volumes across the global crypto market frequently exceed $50 billion to $100 billion depending on volatility cycles, meaning even small vulnerabilities in automated systems can scale rapidly. Researchers emphasize that AI agents interacting with trading APIs, wallet infrastructure, or smart contract interfaces could unintentionally propagate erroneous or malicious instructions across financial systems.

The report also highlights the growing use of AI agents in yield optimization strategies, liquidity provisioning, and arbitrage execution. These systems often operate with limited human oversight, increasing the risk that compromised inputs or flawed model behavior could result in cascading financial consequences.

Security Model Shift: From Trusted to Adversarial Design

A central recommendation from the researchers is a shift in design philosophy: AI agents should no longer be treated as trusted intermediaries but instead as adversarial components within a broader security framework. This includes stricter input validation, sandboxed execution environments, and continuous monitoring of agent behavior.

In traditional software systems, trust boundaries are typically defined at the network or application layer. However, AI agents introduce a new abstraction layer where decision-making itself becomes dynamic and probabilistic, complicating conventional security models.

For crypto infrastructure providers, this shift may require redesigning how automated trading bots, portfolio managers, and DeFi interfaces interact with underlying protocols. Exchanges and blockchain developers are already under pressure to strengthen safeguards as institutional participation increases and regulatory scrutiny intensifies globally.

Investor and Institutional Risk Considerations

From an investor perspective, the growing reliance on AI agents introduces both efficiency gains and systemic risk considerations. Institutional capital increasingly depends on algorithmic execution systems that rely on AI for trade timing, risk management, and market signal interpretation.

While these systems can improve liquidity and reduce execution costs, they also introduce correlated risk if multiple agents respond similarly to market signals or data manipulation. In stressed market conditions, such behavior could amplify volatility across already interconnected crypto markets.

Psychologically, market participants tend to overestimate the reliability of automation during stable conditions, while underestimating systemic fragility during periods of stress. Researchers caution that this imbalance can contribute to delayed risk recognition in highly automated environments.

Outlook for AI Security in Financial Infrastructure

The warning that AI agents must be treated as untrusted systems underscores a broader transition in digital infrastructure security. As AI becomes embedded in crypto trading, blockchain operations, and financial automation, the attack surface expands beyond traditional cyber threats into behavioral and model-level vulnerabilities.

Future development is likely to focus on hardened AI frameworks, improved adversarial testing, and tighter integration between security engineering and machine learning design. For crypto markets, the challenge will be balancing the efficiency gains of AI automation with the need to preserve systemic stability and trust in increasingly autonomous financial systems.