Coinbase CEO Brian Armstrong argued that AI will make software more secure, not less, because security teams can scan code before it reaches production.
“AI will actually make software more secure, not less,” Armstrong wrote on X. He added that the technology favors defenders over attackers because defenders can inspect an entire codebase before release, while attackers still need exploitable gaps to remain unnoticed.
The comment pushes back against a growing fear that AI will create a one-sided advantage for hackers. In Armstrong’s view, the same tools that can help attackers search for bugs can also let software teams run massive automated reviews, detect vulnerabilities earlier and catch issues that human reviewers may miss.
The argument lands at a sensitive moment for crypto. Coinbase has been moving AI deeper into financial workflows through Coinbase for Agents, a product that lets AI assistants connect to user accounts, trade and prepare for stablecoin-powered payments inside permissioned limits.
Crypto Security Debate Turns To AI
Armstrong’s point fits a wider defender-advantage view that has long existed in cryptography. Ethereum co-founder Vitalik Buterin has described cryptography as one of the few fields where adversarial conflict can heavily favor the defender, because strong keys and well-designed systems can be far more expensive to break than to maintain.
AI now brings that debate into software development. Smart contracts, wallets, bridges, exchanges and AI-agent accounts all depend on code paths that can move money instantly. A stronger pre-production review process could reduce basic bugs, permission mistakes, unsafe dependencies and logic flaws before attackers reach them.
The risk is that AI also helps attackers move faster. Security agencies have warned that AI is accelerating cyber risk by lowering the skill barrier for attack tooling, vulnerability discovery and automated exploitation. That makes patch speed, dependency monitoring, permissions and production controls more important, not less.
Crypto has already seen the financial version of that problem. A Grok-linked wallet drain showed how AI-connected accounts can become exposed when tool permissions, wallet execution and prompt handling are not tightly controlled.
AI Agents Raise The Stakes For Code Review
The security question is becoming more urgent because AI agents are starting to transact. Coinbase’s x402 protocol has already processed 160 million agentic payments, showing that machine-driven payment flows are moving beyond demos and into real usage.
Once agents can trade, pay, call APIs, access wallets or interact with DeFi, software flaws become direct financial risks. A weak approval flow, unsafe tool permission, bad dependency update or missed validation rule can expose assets before a human notices the problem.
Armstrong’s defense-focused argument depends on how teams actually use AI. The strongest version would combine AI code scanning, formal verification, human review, audit trails, sandboxed permissions, limited spending rights and production monitoring. The weaker version would rely on AI to write and approve code without enough independent security review.
