A $6 billion hack in a season of AI anxiety

November 3, 2026 • 5:44 pm ET JP Schnapper-Casteras In a scene from The Wizard of Oz, Dorothy and her companions follow the yellow brick road through a dark forest, filled with frightening noises. As panic mounts, they repeat the famous line: “Lions and tigers and bears—oh my!” Last week had a similar vibe. There was a $6.3 billion hack of cryptocurrency exchange Binance and artificial intelligence (AI)–assisted discoveries of major vulnerabilities in a programming language that underpins much of the internet. Those headlines were understandably unsettling, particularly layered atop periodic predictions of an AI apocalypse. But to borrow a lesson from Dorothy, not everything in the forest is as new or terrifying as it might seem upon first glance. Some of the creatures may be familiar. Distinguishing genuinely new threats from old problems—and changes in their speed, scale, or consequences—should guide how financial institutions and policymakers respond. A familiar kind of heist The Binance hack is both somewhat new and mostly familiar. Here’s what happened in a nutshell: On October 29, attackers stole more than six billion dollars in digital assets from the exchange. The Federal Bureau of Investigation (FBI) has attributed the theft to North Korea’s TraderTraitor operation. So far, there is no public evidence that AI tools or “swarms” were used to break into Binance’s systems. Binance’s initial investigation suggested an engineer’s compromised laptop, but later identified additional points of breach. And as former PayPal executive David Marcus observed, compromised credentials and trusted access are familiar weaknesses in financial security. Nor are major hacks of cryptocurrency exchanges anything new: the FBI attributed last year’s $1.5 billion Bybit theft⁠ to the same North Korean operation, while Mt. Gox (2014), Bitfinex (2016), Coincheck (2018)⁠, and Binance itself (2019)⁠ had previously suffered major breaches. What happened afterward is more striking and potentially novel. Chainalysis identified thousands of transactions moving funds repeatedly among digital wallets. Some commentators have suggested this could be an early, high-profile example of “agentic money laundering.” But the patterns suggest automation, not necessarily the involvement of AI agents. What may be changing more dramatically is the speed at which stolen funds can move, leaving less time to freeze assets or block suspicious transfers. Our investigators are observing never-seen-before laundering patterns with the stolen funds from the Binance exploit. Thousands of transactions in a short period suggest that the attackers are using automation to obscure the trail of funds. Rather than simply moving funds onward,… pic.twitter.com/Uhu3dbxDqh— Chainalysis (@chainalysis) October 30, 2026 There are also reasons not to despair. The FBI, Secret Service, Internal Revenue Service, blockchain analysis firms, and private companies have worked to trace and freeze stolen assets. Binance reports that it has replenished about 95 percent of the affected assets using its own resources—although that is not the same as recovering 95 percent of the money from the hackers. Looking ahead, the challenge is bigger than cryptocurrency. The Financial Action Task Force has recently documented how professional money-laundering networks exploit both traditional and newer financial channels. And as the International Monetary Fund has warned, agentic payments may complicate questions about who authorized a transaction, what an agent was permitted to do, and how suspicious activity should be monitored. A new vulnerability race Next, the headlines around the use of AI in unearthing critical software vulnerabilities are, on their face, more novel and troubling. On October 21, researchers at Wiz, which Google acquired earlier this year, disclosed serious flaws in PHP, a widely used open-source programming language, uncovered with the help of frontier AI models. Separately, Node.js patched a severe vulnerability reported by OpenAI. Discovering a vulnerability, of course, is not the same as exploiting it—and security researchers have long disclosed software flaws and published related technical analysis. What may be new is the speed with which increasingly powerful AI models can help researchers—as well as malicious actors—find and test longstanding weaknesses. This is likely to be the beginning of a tumultuous period for chief information security officers and software maintainers as they race to identify and patch weaknesses—and face potential attackers armed with similar tools. But there are several reasons for guarded optimism: Wiz privately reported the PHP vulnerabilities in August, nearly two months before public disclosure and the release of patches in October. That gave maintainers and service providers time to prepare patches and mitigations, although leaks in the final days complicated the coordinated release. Additionally, major financial institutions and technology companies are already testing powerful defensive AI tools. JPMorganChase, for example, is a launch partner in Anthropic’s Project Glasswing, which provides early access to its Claude Mythos model for defensive security work. A KPMG survey found that 84 percent of banking executives were increasing cybersecurity investment to address AI-related risks. That does not necessarily mean all are investing in defensive AI tooling, but it illustrates the changing priorities. More broadly, the benefits of defensive AI should extend beyond Wall Street and Silicon Valley, particularly to operators of critical infrastructure (such as electric and water utilities, which may rely on aging software and control systems) and maintainers of open-source software projects (which often have more limited cybersecurity resources). Just this week, cyberattacks hit supermarket refrigeration systems across Europe, a reminder that software vulnerabilities can have tangible consequences in the physical world. There may be a larger opportunity here, too: using AI not simply to discover and patch one dangerous bug at a time, but to proactively examine and repair (or refactor) aging code that underpins the internet. The same tools that could help attackers may also help defenders get there first. Old laws, novel questions For legislators and litigators, the bigger takeaway from these two trendlines is that the next five to ten years may usher in a host of complicated legal questions at the intersection of cybersecurity, AI, and financial technology. For example, when might an AI company face liability for the second-order consequences of its tools or agents? And what happens if and when an autonomous agent causes harm without explicit human direction? Some problems may require new rules or regulations. Others will be settled by laws and principles already on the books. As a recent Congressional Research Service analysis explains, existing federal criminal laws can apply when a person deliberately uses AI to gain unauthorized access to a computer, even if unintended autonomous conduct raises harder questions. Even long-established principles of negligence and comparative fault may be well suited to sorting out responsibility among AI firms that disseminate dangerous vulnerabilities, financial institutions that fail to address them, and other parties whose actions contribute to a loss. At bottom, there are surely scary twists and turns ahead as we move toward increasingly automated finance. But much like Dorothy, we may discover that some of the lions are old, familiar creatures—even if they pose some risks. Regulators and policymakers might save themselves considerable time and anguish by distinguishing genuinely new problems from familiar misconduct, and concrete risks from conjecture, before reaching for sweeping new restrictions. The challenge is not simply to brace for some conjectured AI apocalypse, but to focus on the real risks emerging today and discern what is new, what is not, and how best to respond. JP Schnapper-Casteras is a nonresident senior fellow at the Atlantic Council’s GeoEconomics Center and general counsel and chief compliance officer of Collective Global, a venture capital firm with investments in the technology and AI sectors. At the intersection of economics, finance, and foreign policy, the GeoEconomics Center is a translation hub with the goal of helping shape a better global economic future. Image: Binance logo and hacker code are seen in this illustration taken October 30, 2026. REUTERS/Dado Ruvic/Illustration