Web browser icon with series of lines on an orange background.
Web browser icon with series of lines on an orange background.

AI-Driven Attacks Are Here. Five Steps CISOs Should Take Now

Arctic Wolf case data shows network intrusions rose 96% as attackers adopt AI. Learn five steps CISOs can take now to speed response and build resilience.
Web browser icon with series of lines on an orange background.
6 min read

Earlier this year, Claude Mythos Preview showed the security community what a frontier AI model can do when it’s pointed at software, finding and linking vulnerabilities far faster than defenders had ever seen up to that point. Access to that model stayed tightly controlled, but ARTEX has recently removed that corporate barrier for attackers. This open-source attack framework works with almost any LLM, and threat actors began experimenting with it against organizations within weeks of its release.

Although the developer later removed the project from GitHub, the code was quickly mirrored elsewhere and remains accessible. More importantly, we’re seeing these techniques attempted against customers. Arctic Wolf is tracking the emerging use of self-hosted agentic penetration testing frameworks like ARTEX, including AntiFox, operating from public cloud infrastructure. In several recently observed intrusions, Alibaba Cloud systems hosting AntiFox interfaces were linked to confirmed malicious activity targeting internet-facing web services and associated AWS environments.1 There is no indication the hosting provider was aware of or involved in this activity.

We’re seeing the effects of these enhanced offensive AI capabilities in our own case data. In what some industry leaders are calling ‘Black October’ due to the spike in vulnerability disclosures and threat activity, Arctic Wolf Aurora Incident Response handled 78% more non-Arctic Wolf customer cases than the average month prior, and although Business email compromise (BEC) activity remained relatively flat, more complex attacks surged. Network intrusion attacks, including ransomware, rose by 96%, meaning that attackers aren’t just more active. They’re more dangerous.

CISOs know AI is changing the threat landscape, and our case data reflects that shift. To adapt their security operations to keep pace with this shift, security teams should focus on five areas.

Reset Your Response Timelines

Most incident response plans created pre-AI, or even prior to this year, assume there will be enough time to assess what’s happening, gather the right stakeholders, and make containment decisions before an attacker causes significant damage. Increasingly, that’s not the case, especially in large organizations with long chains of approval. Because AI is helping attackers move faster, organizations need to remove as much friction from their response plan as possible, which means deciding in advance how a security team can take action without waiting for approval.

Map Everything Attackers Can See

AI-driven tools can scan thousands of potential targets at once and quickly identify the easiest path into an environment. That’s why there’s never been a better time for defenders to take full inventory of what systems, applications, and services they use are exposed to the internet. With a full inventory, security teams can prioritize taking action on the exposures that are likely to be targeted first.

Strengthen Identity Controls

Social engineering and external remote access ranked among the most common attack vectors in our October engagements, and AI is making both easier to execute at scale. An attacker can now generate extremely convincing phishing messages, and thus targeted campaigns, in seconds using publicly available information. That makes strong identity controls absolutely critical in today’s security landscape. Organizations should prioritize phishing-resistant multi-factor authentication, regularly review their remote access tools and old or inactive accounts, as well as apply the same level of scrutiny to their service accounts and API keys. Those often carry extensive permissions but receive far less oversight than human users, making them an attractive target for attackers.

Monitor Behavior at Runtime

Some malware now contacts an LLM during execution to retrieve instructions on demand, which means some of the script may not exist until the malware is already running. The problem is that traditional controls can miss these kinds of attacks, and defenders may not be paying close attention to their AI services. These signals can be correlated across endpoints and network activity, but they usually occur outside of normal business hours, so organizations need the ability to investigate and respond continuously with 24×7 detection.

Practice Before You Need It

Teams that have already worked through their response process generally spend less time figuring out roles, decisions, and communications when a real incident occurs. Leaders need to regularly schedule tabletop exercises built around AI-driven scenarios, especially as new offensive capabilities go live, and test how their team handles a fast-moving attack, like a ransomware precursor or a large data theft. Security leaders also need to confirm that they have an incident response partner ready to engage quickly when something happens.

Building Resilience at Machine Speed

The fundamentals of security haven’t changed. Organizations still need strong identity controls, visibility into their environment, tested response plans, and the ability to detect suspicious behavior quickly. What has changed is the speed at which attackers can operate. As AI continues to compress the time between compromise and impact, security teams need to make sure their people, processes, and technology can keep up.

That’s why Arctic Wolf has built the Aurora Superintelligence Platform and the Aurora Agentic SOC. Our approach has always been grounded in helping organizations operationalize the fundamentals of security. That means maintaining visibility across their environment, strengthening identity security, continuously monitoring for signs of compromise, and ensuring skilled responders are ready when an incident occurs.

The addition of AI and agentic capabilities allows us to help customers execute those tasks faster and at greater scale, and our 2026 Agentic Security Operations Report outlines the impact we’ve measured. Over the past year of the Aurora Agentic SOC, investigation volume has increased 21%, while alert volume has been reduced by 19%, and we’ve shortened the time from an alert-generating event to an actionable ticket by 36%.

At the same time, agentic AI now resolves 60% of investigations, allowing human analysts to provide oversight when necessary. As attackers increasingly use AI to identify exposures, move laterally, and accelerate their operations, we’re delivering the technology designed to help defenders rapidly understand their environment, prioritize risk, investigate suspicious activity, and take action before a threat turns into a business disruption.

The future of security operations will be defined by how effectively organizations combine machine-speed analysis with human oversight. Organizations that can pair strong security fundamentals with trusted AI-driven operations will be best positioned to reduce risk, respond faster, and build resilience against the next generation of threats.

1 Attribution reflects Arctic Wolf Labs’ assessment as of the report period and may evolve with new evidence. References to threat actor identity, nexus, and intent are analytical judgments, not statements of legal fact.

Share this post: