CertiK CEO Warns Mass Deployment of AI Agents Poses Severe Security Risks
AI agents can autonomously access accounts, sensitive data and blockchain wallets. If deployed at scale without security reviews or isolation, a vulnerability in a single model could spread across an entire system. Ronghui Gu, CEO of blockchain security company CertiK, said such agents could put credentials and digital assets at risk, making zero-trust architecture a critical line of defense.
In the latest report as of July 20, 2026, Gu warned that hackers could use prompt-injection attacks to manipulate AI agents into exposing login credentials, executing unauthorized transactions or participating in onchain fraud. He called for code audits, permission isolation and continuous monitoring before deployment. The report disclosed neither actual losses nor the scale of any confirmed attacks.
All Coverage
1 original reportsThe Backstory
The history behind this eventAI Agents Threaten to Supercharge Crypto Hacks, Industry Leaders Warn
AI agents can plan tasks, invoke software tools and authorize onchain transactions rather than merely offer advice. That autonomy could improve trading and payments, but it also lets prompt injection, hallucinations or malicious instructions propagate at machine speed. Because blockchain transfers are difficult to reverse, the industry must determine how agents receive wallet permissions, how their actions are audited and whether users, developers or service providers bear responsibility when losses occur.
Industry leaders warned in August 2026 that attackers could deploy fleets of agents to scan targets continuously and steal small amounts every day, potentially making today’s billion-dollar crypto heists look trivial. In December 2025, Anthropic tested leading models against 405 previously exploited smart contracts; the agents uncovered $4.6 million of exploitable value, while an exhaustive scan cost just $1.22 on average. Executives said unresolved trust, model hallucinations and legal liability remain the biggest barriers to widespread adoption.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →