Phemex CEO: AI ‘net negative’ for crypto, fueling attacks and costs

Phemex CEO Federico Variola has taken a critical stance on AI’s growing role in crypto, arguing that the technology is currently acting more like a drag than a catalyst for the sector. In comments on Cointelegraph’s Chain Reaction, Variola said AI has redirected liquidity away from crypto and intensified security risks—especially for smaller teams that cannot easily absorb rising defense costs.

While Phemex earlier this year outlined an AI-led shift in its own product and internal workflows, Variola’s latest remarks expanded the focus to the wider industry. He suggested the net effect of AI may be to push crypto further toward centralization rather than strengthening decentralization.

Key takeaways

• Variola argues AI is diverting capital from crypto to other sectors while simultaneously increasing the power of threat actors.
• He warns that higher cybersecurity burdens may concentrate security capacity among well-funded entities, potentially encouraging centralization.
• Recent crypto incidents have been linked by industry figures to AI-enabled tactics, raising the perceived baseline risk for both self-custody and DeFi.
• Even so, Variola sees practical uses for AI agents in portfolio support and decision-making, while stopping short of expecting full replacement of human judgment in trading.

Why Variola says AI is a “net negative”

Speaking during the episode, Variola said it is hard to be bullish about AI within crypto because it changes incentives on both sides of the equation. According to him, liquidity has been significantly rerouted into AI-related activity, reducing the competitive attention and funding available to the crypto industry itself.

At the same time, Variola argued AI is accelerating adversarial capabilities. In his view, AI is not simply raising the technical quality of code—it is also helping attackers exploit protocols and scale tactics that rely on automation, speed, and sophistication.

“It’s difficult to envision a world in which AI is going to favor crypto specifically as an industry, since a lot of the fixes that we see actually encourage more centralization rather than less centralization.”

For investors and builders, the core tension in Variola’s argument is straightforward: if staying safe increasingly requires large security budgets, then the economic pressure to centralize operations may intensify. That matters for an industry whose claims to value often depend on decentralization and broad participation.

Security pressure as AI changes the threat landscape

Variola’s skepticism lands alongside a period in which AI has repeatedly entered discussions around crypto security incidents. One example cited in the broader conversation is a July incident in which attackers reportedly drained roughly $116 million in Bitcoin from more than 5,200 addresses tied to a Coldcard hardware wallet flaw. The report framework around that incident has widely been tied—at least in industry narratives—to the malicious use of AI.

Coinkite CEO Rodolfo Novak also publicly warned developers at the time that the “sober reality” is that AI-assisted code review can uncover bugs faster than even experienced teams can keep up with. That perspective aligns with Variola’s concern: the same speed and scale that makes AI useful can also shorten the window between vulnerability discovery and exploitation.

Variola explicitly tied these risks to the economics of security staffing. He said small protocol teams may no longer be able to function effectively without “a massive cybersecurity budget,” implying that defensive measures may become structurally out of reach for many decentralized projects.

The stakes are not limited to protocol-level vulnerabilities. Variola also suggested that AI-driven threats could reduce the appeal of self-custody and decentralized finance for everyday users. He argued that as AI becomes more pervasive, attacks may increasingly take the form of device compromises or social engineering campaigns—conditions that could raise the number of risks retail users must actively manage.

“As AI becomes more pervasive, and whether it is your devices being hacked or social engineering, or all these kinds of strategies that are empowering threat actors, that makes DeFi a lot less appealing for a retail user because you have to worry about so many things that you didn’t have as much before.”

Defensive AI exists—but the trade-off may still be hard

Not everyone in the security community shares the same level of pessimism. The article notes that security experts have also highlighted AI’s defensive potential. For example, CertiK senior blockchain investigator Natalie Newson told Cointelegraph in April that “AI can also be one of the biggest defenses,” even while warning that attackers are simultaneously becoming more sophisticated.

This is a critical nuance for readers: the debate is not simply whether AI helps attackers or defenders. It is about pace, cost, and who can afford to respond quickly. If attackers can iterate faster while defenders face rising operational expenses, the overall balance may still tilt toward the party with resources—whether that is a centralized service, a large security team, or an entity able to purchase rapid incident response.

That framing supports Variola’s larger thesis: even if AI improves security tools, the resulting “fixes” may require organizational scale that decentralized projects struggle to replicate.

Where Variola sees room for AI in crypto

Despite his broader caution, Variola said he sees tangible benefits in AI agents—specifically for user-facing assistance such as helping investors build portfolios or improving trading decisions. However, he emphasized that these tools should be treated as decision support rather than autonomous replacements.

In his words, users would remain responsible for the final action: AI agents might help with information processing, strategy framing, or execution context, but they are not expected to fully replace the judgment involved in taking a trade.

For market participants, this distinction may be practical. Portfolio guidance and workflow automation can reduce friction for retail users, but security and risk management still depend heavily on human oversight—especially in a threat environment Variola describes as increasingly shaped by social engineering and device compromise.

As AI adoption accelerates across crypto platforms, readers should watch two closely connected questions: whether AI-enabled defenses meaningfully reduce real-world exploit frequency, and whether the cost of those defenses pressures the ecosystem toward larger, more centralized operators. The answers will likely determine whether AI ultimately broadens participation—or quietly narrows it.