Black October Underscores a New Era of Cyber Offense and Defense

October is Cybersecurity Awareness Month. But this past October, cybersecurity developments brought some digital weaknesses to the surface, raising concerns across the industry and among policymakers. The systems Americans have come to rely on every day, and take as relatively secure, are starting to show some serious weaknesses.

The lesson is becoming clearer by the day: nothing—even the most promising technologies—is completely immune to hacks, leaks, and other data incidents. Private companies have strong incentives to respond to and patch up these vulnerabilities—so long as policymakers do not prevent them from doing so. However, these threats would be increased by legislation, such as the American Innovation and Choice Online Act, that would weaken companies’ ability to safeguard their consumers’ cybersecurity. Moreover, mandates forcing the collection and aggregation of massive amounts of information, such as age verification mandates, multiply the risks of any of these vulnerabilities. These tradeoffs must be weighed, and with every new data breach, the costs of enacting privacy-impairing policies become more apparent.

Throughout October, researchers uncovered vulnerabilities and reported cyberattacks affecting companies across different sectors. Notably, finding weaknesses in complex systems has traditionally required significant technical expertise, time, and resources, but artificial intelligence (AI) is giving researchers new ways to detect flaws buried in complex systems—a process that once could take months or even years.

AI is lowering the barriers to both discovering and exploiting vulnerabilities by changing the speed, scale, and sophistication of cyber operations. But while these capabilities allow researchers to identify vulnerabilities, they can also be used by malicious actors to exploit them. These advances are happening while longstanding weaknesses—including compromised credentials, vulnerable software, and insecure third-party services—in the software and digital infrastructure continue to threaten the applications and devices people use every day.

Cyberattacks were reported on October at an unusually high rate, using different methods and targeting different parts of the digital ecosystem. CNBC reported that roughly 3 percent of all critical vulnerabilities ever discovered turned up in the past month alone, a sign of how AI models are making vulnerability research much easier. But identifying vulnerabilities is only part of the challenge. Once a vulnerability becomes public, attackers may move quickly to exploit it, leaving organizations with less time to respond.

A report by Wiz found that nine AI developer platforms reported security incidents between October 6 and October 30. While the incidents varied in method, several involved a technique known as LLMjacking—in which attackers steal or abuse credentials to access someone else’s AI models and infrastructure—while others involved more sophisticated techniques, including attempts to extract or replicate the capabilities of AI models through a process known as model distillation. A separate campaign, reported by Reuters on October 19, involved a hacker who used coordinated AI agents to try to break into more than 1,200 websites across the United States, Canada, Australia, and Japan. More than 100 of them were compromised. By combining DeepSeek with Anthropic’s Claude Code, the attacker used AI agents to find vulnerabilities and carry out intrusions on their own, with minimal human oversight.

Another similar case was the discovery of vulnerabilities in PHP, a programming language used by approximately 70 percent of websites worldwide. Researchers at Wiz identified flaws in PHP’s core engine that had gone undetected for years. If remained undetected, these flaws could allow attackers to remotely execute malicious code on vulnerable web servers. Other reports soon documented attempts to exploit the flaws, showing how quickly a discovery made by security researchers can become an opportunity for attackers.

That same week, the software of LiteLLM, an AI infrastructure platform, was compromised after a malicious version of its software was distributed online. The compromised version contained malicious code that could run automatically, even when developers were not actively using LiteLLM. Unlike the previous incidents, this was not an AI-powered attack, but a more traditional software attack. Still, it showed how compromising a single tool used by developers can expose sensitive information and create risks for other connected systems.

In another report published on October 22, researchers also found a flaw in RingRTC, the technology Signal uses for voice and video calls. It could allow someone to keep a call active even after the other person had hung up, potentially leaving their microphone or camera on. While the vulnerability was fixed quickly, the incident shows how vulnerabilities in underlying software components can compromise user privacy, even in applications specifically designed to protect communications.

Additionally, these incidents have brought attention to how interconnected the security of the various parts of digital systems really is. From the tools developers use to build software to the cloud infrastructure, ad networks, and third-party services that applications rely on, a vulnerability at any layer can ripple across the entire ecosystem. Protecting it will therefore require shared efforts from developers, technology companies, infrastructure providers, organizations, and policymakers to strengthen security across the software lifecycle and put user privacy first. This can best be accomplished through a light-touch approach, not a heavy-handed one. Innovators are poised to respond to the latest risks and will likely succeed—unless ill-conceived mandates thwart their efforts and make the situation far worse.