The rapid evolution of artificial intelligence, particularly advancements from China, is raising stark concerns among global cybersecurity professionals. During a recent discussion at a Washington D.C. conference, Netskope CEO Sanjay Beri cautioned about the potential implications of cutting-edge AI-driven cyber threats, stressing the urgent need for companies to invest in robust defense systems.

Beri spoke in light of revelations about Claude Mythos, a highly advanced AI model under development by Anthropic. The model, designed to identify and manipulate vulnerabilities, has exceeded developmental benchmarks, prompting careful deliberation over its release. The model’s capabilities are so potent that Anthropic created the 'Glasswing' initiative—a controlled deployment to test its functionality with selected major corporations. Among those involved were tech behemoths like Microsoft, Google, and Cisco, alongside major financial institutions, with Federal Reserve Chair Jerome Powell and key banking CEOs reportedly being briefed on its cybersecurity implications.

“We have been developing [Claude Mythos], and it blew every benchmark we had out of the water,” Anthropic cofounder Jack Clark stated. However, the power of such technology cuts both ways. If exploited, these advanced AI models could be used maliciously, finding and combining system vulnerabilities faster than ever before. While this capability offers new tools for defenders, it equally arms attackers if they manage to access similarly advanced AI systems.

The potential from Chinese AI

A significant portion of the conversation highlighted concerns about China’s swift advancements in generative AI. When asked about the risks presented by Chinese models matching or surpassing Claude Mythos, Beri gave a firm warning: “Never underestimate Chinese models. They have come very quickly, and [China has] a history of acquiring models quickly—bluntly, even stealing them.” Beri noted that it is not unreasonable to assume Chinese entities possess comparable capabilities already, given their aggressive investments and focus on AI.

He explained that things that once took months for attackers to uncover manually can now be automated and executed in mere hours, thanks to sophisticated AI. Such advancements drastically reduce the timeframe for identifying systemic vulnerabilities and building exploits, creating significant challenges for organizations trying to secure sensitive data. The risk to intellectual property, financial records, and other confidential information is immense.

The corporate perspective on AI risks

Beri outlined how corporate leaders increasingly recognize the stakes. At a recent security conference where he spoke to over 100 chief information officers, concerns over managing internal AI risks were a recurring theme. Companies are not only worried about external attackers but also about their internal exposure when employees use generative AI tools naively. Simple actions, like inputting proprietary information into external tools like ChatGPT, could inadvertently leak sensitive company data.

“A lot of companies are asking us for help in protecting their AI usage,” Beri said. Netskope specializes in overseeing and mitigating these risks. According to him, tools from companies like his enable monitoring and controlling user interactions with generative AI. For instance, if an employee attempts to process healthcare data using ChatGPT or a rogue program threatens to expose sensitive financial information, Netskope’s platform can identify and neutralize the potential breach in real-time. This proactive approach prevents users, agents, or automated systems from engaging in harmful activities, securing the organization’s data integrity.

A broader shift in cybersecurity

For cybersecurity as a field, the rise of advanced AI represents both an opportunity and a challenge. Beri emphasized that the industry could no longer view threats in isolation. Cybersecurity must operate on a principle of assumed compromise with layered defenses, a concept often called 'defense in depth.' Organizations need to anticipate vulnerabilities rather than merely react to breaches once they occur.

Despite Netskope’s expertise in this domain, challenges remain on the financial and operational front. Beri defended the company’s long-term prospects and approach amidst investor skepticism over cybersecurity stocks, some of which have seen sharp declines. He attributed part of this to a lack of differentiation between traditional software and highly specialized cybersecurity firms like Netskope. While acknowledging competition from legacy vendors like Broadcom, Beri underscored Netskope’s focus on addressing modern risks tied to cloud platforms and AI-driven threats. “We’re built for the cloud and AI world… Our job is making sure enterprises worldwide know who we are,” he stated confidently.

Why it matters

The stakes surrounding AI-driven cybersecurity are rising rapidly. Tools like Anthropic’s Claude Mythos or open-source models informed by Chinese development are game-changers in the cybersecurity landscape. While they hold promise for advancing protections against malicious exploits, their power in the hands of bad actors could be catastrophic. Beri’s warning underscores that this issue is no longer hypothetical—companies must act now to adapt to this new reality.

Part of the solution lies in cultivating a deep understanding of AI risks and deploying appropriate monitoring tools. But the broader challenge will be staying ahead of increasingly complex and global threats in an era where vulnerabilities could be developed and exploited at unprecedented speeds. The question is no longer whether AI will reshape the cybersecurity fight—it has. What matters now is ensuring defenders are equipped as well as, if not better than, potential adversaries.

With China’s AI capabilities advancing rapidly and major warnings from leading executives, the tech world must focus on finding innovative ways to protect sensitive systems in an increasingly uncertain environment.