Microsoft co-founder Bill Gates said in a mid-October CNN interview that the most significant changes in the year ahead are likely to come from cyberattacks—particularly those enabled by advances in AI. Gates warned that developers can switch off safety features in AI models and that the primary threat is how malicious actors can exploit those weaknesses. The comments arrive as industry data points to fast-growing AI-driven attack activity, with CrowdStrike reporting an 89% year-over-year increase in AI-powered cyberattacks in 2025.
Key takeaways
- Price move: No specific market price move was reported in the source material.
- Catalyst: Bill Gates’ warning about AI-enabled cyber threats, alongside cited findings on rising AI-powered attacks.
- Key implication: Investors and consumers may need to reassess how much personal and financial information they disclose to AI tools, because risk can come from what can be inferred from combined details.
- Security focus: Data minimization is becoming more critical as AI assistants shift from one-off queries to persistent systems that remember user preferences and context.
What drove the warning about AI and cyber risk
Gates’ central point was that AI safety controls are not immutable. He noted that developers can disable built-in protections in AI models, leaving room for misuse. He framed the danger less around the existence of AI itself and more around how it can be reconfigured for harmful ends—especially when attackers can prompt models to generate content, facilitate scams, or support social engineering.
Coverage of Gates’ comments also referenced CrowdStrike data showing an 89% year-over-year increase in AI-powered cyberattacks in 2025. While the reporting did not specify which attack categories drove that growth, the implication for users is that adversaries are increasingly using AI to scale operations, personalize lures, and reduce the friction of launching targeted fraud.
Why “harmless” conversations can become a roadmap for fraud
Vugar Usi, head of major crypto exchange MEXC, argued that the risk profile from AI interaction is shifting. He suggested that the most sensitive information—such as passwords or seed phrases in the crypto context—is often already protected by experienced users. Instead, he highlighted the danger of disclosures that appear benign when shared in isolation.
Usi described how a fraudster could build credibility through a sequence of conversations: discussing a dog’s name, mentioning losses from perpetual futures, asking how to withdraw funds from a particular exchange, and then moving to topics like a mortgage or a child’s education expenses. No single detail is necessarily alarming, but the combination can help a scammer craft a convincing narrative and pressure the target into acting quickly.
The concern, Usi said, is that modern fraud does not always need generic openings. Rather than starting with a “Dear Customer” style approach, attackers can use context-rich messages that make it appear they are already tied to a real account or a specific, genuine problem. Data exposure and data linkage make this easier: names, email addresses, and phone numbers frequently surface in leaked databases, while the history of AI interactions can add missing context that helps connect separate details into a more complete financial profile.
How widespread AI usage raises the stakes
Usi pointed to survey and research findings to argue that these risks are becoming mainstream rather than niche. According to a 2024 Experian survey cited in the article, nearly half of American consumers have used or considered using generative AI to manage personal finances. The share was higher among younger cohorts, reaching 67% for Gen Z and 62% for millennials.
Within crypto, the article cited a CoinGecko survey indicating that 35.9% of respondents would be comfortable entrusting 60% or more of their portfolio to an AI agent, and about one in seven would hand over their entire portfolio. Separate research cited in the article also suggested that users are already sharing confidential inputs: a 2024 Cisco study found 30% of generative AI users entered personal or confidential information into these systems, including 29% sharing financial details and 27% providing account numbers.
Further, the piece referenced workplace-focused research by Harmonic Security covering more than 22 million prompts and file uploads in 2025. The analysis reported over 579,000 instances involving sensitive data exposure. It also noted that sensitive material appeared in approximately 4.4% of prompts and 22% of uploaded files.
Bigger picture: what changes when AI becomes a persistent assistant
Usi emphasized that files and documents may be an especially overlooked risk. Even if a tax return removes direct identifiers, other details—such as employer, income, and investment information—can still enable identification. He also argued that AI tools often do not need highly granular data like account numbers or precise balances to perform portfolio-related analysis, meaning users can reduce exposure without sacrificing utility.
The article also cautioned against assuming that AI services themselves are inherently secure. It referenced a January 2025 incident reported by Wiz in which an exposed DeepSeek database reportedly contained more than a million rows of logs, including chat histories and secret keys, accessible without authentication until the vulnerability was closed after notification.
While Usi said AI assistants are not necessarily more dangerous than other online services—and noted that banks and government agencies have also suffered breaches—the core principle he stressed was data minimization. As AI systems evolve from tools for one-off questions into persistent companions that retain preferences and context, the potential upside for users comes with a larger footprint of personal and financial data—making it more important to avoid providing information a system does not truly require.
In that context, the article suggested that classic crypto security guidance such as “not your keys, not your coins” should be complemented by a simpler rule: do not share what a machine does not need to know. The underlying message for consumers and market participants is that the risk may increasingly be determined by inference—what can be deduced from the combination of small details—rather than by any single sensitive item alone.
What to watch next: With AI-driven cyberattacks accelerating, investors and industry watchers will likely focus on whether companies tighten AI safety controls, improve data handling policies, and strengthen protections around sensitive inputs—especially user files and conversation history. Upcoming catalysts include continued reporting on AI-related breaches and any further regulatory or corporate guidance on safe AI usage.
