AI Agent Security Concerns Grow as Researchers Warn Against Blind Trust

Artificial intelligence agents are gaining momentum across crypto and digital finance, but new research warns that these systems should not be trusted by default. Security experts argue that AI agents must be treated as untrusted systems to reduce risks tied to hacking, manipulation and unauthorized access.

AI Agent Security Requires Stronger System Protection

A research paper updated on May 20 by experts from Google, Gray Swan AI, EmbraceTheRed and several universities said AI security should be handled as a full systems problem rather than focusing only on improving model behavior.

Researchers argued that strengthening AI models alone is not enough to stop attacks. Instead, developers should rely on proven computer security methods that have long protected systems against powerful threats.

The report recommends separating trusted instructions from untrusted data, reducing the chances of attackers hiding harmful commands inside ordinary information. Researchers also stressed that AI agents should operate with minimum permissions, limiting access only to tasks required for completion.

Crypto Industry Sees Rising Use of AI Agents

AI agents are becoming more common in crypto trading, Web3 applications and automated blockchain services. Circle CEO Jeremy Allaire recently predicted that billions of AI agents could operate on behalf of users within five years.

Recent incidents have raised concerns. On May 20, the AI-powered crypto assistant Bankr paused transactions after attackers reportedly accessed at least 14 wallets. Security specialists said stronger controls around wallet permissions, transaction limits and real-time monitoring remain essential as AI adoption expands.