Every business leader who has deployed AI tools in the past year needs to read this. Check Point Research released its Annual AI Security Report 2026 this week, and the headline is not subtle: AI has stopped assisting cyberattacks. Now it is running them.

That is not a theoretical warning. It is a documented shift, backed by real breach data, that every organisation using AI tools needs to understand.

What the Report Actually Found

The report, published July 14-15, 2026, documented intrusions where AI ran complete exploitation workflows with minimal human direction. In one of the more alarming examples, a single operator used Claude Code and GPT-4.1 to breach nine Mexican government agencies, generating 5,317 AI-executed commands across 34 attack sessions. The human operator set the objective. The AI did the work.

This matters because it fundamentally changes the risk equation. Traditional cyberattacks required skilled human operators at every stage. An attacker had to understand the target system, find vulnerabilities, write exploits, and adapt in real time. That took time and expertise, which created a natural ceiling on the scale and speed of attacks.

AI removes that ceiling. When an AI agent can autonomously navigate a network, identify weaknesses, and execute multi-stage intrusions, the number of skilled attackers in the world becomes irrelevant. The bottleneck shifts from human expertise to AI access, and AI access is now cheap and widely available.

The Numbers That Should Concern Every Business

Check Point’s report goes beyond the headline case to paint a broader picture of how AI-assisted threats have scaled in 2026:

• Malicious prompt-injection payload detections rose roughly fivefold between March and May 2026 alone
• High-risk enterprise AI prompts doubled, from about 1 in 50 interactions to 1 in 25
• Between 87% and 93% of organisations now see at least one high-risk AI interaction every month
• AI-driven exploitation workflows are compressing vulnerability-to-exploit time from days to hours

That last point deserves emphasis. The window between a vulnerability being discovered and it being weaponised used to be measured in days, sometimes weeks. Security teams had time to patch. Now that window is measured in hours, and in some cases less. The old assumption that you can patch before attackers weaponise a flaw is no longer reliable.

The Deepfake Problem Is Getting Worse

The report also documents accelerating problems with AI-generated identity fraud. Voice, face, documents, and real-time video can now be convincingly synthesised, and the tools to do this are not confined to nation-state actors. Highly trained human reviewers correctly detect AI-generated faces only 41% of the time.

For businesses that rely on voice or video verification, that number is alarming. For businesses running AI voice agents or AI-assisted customer interactions, it creates new exposure on both sides: your customers can be deceived by attackers impersonating your systems, and your systems can be deceived by attackers impersonating your customers.

Why This Is Especially Relevant Right Now

The timing of this report matters. Enterprise AI adoption has accelerated sharply through 2026, with 72% of agent-based AI now in production and 40% of enterprise applications forecast to incorporate AI agents by year-end. Most organisations moved fast to deploy. Many did not move equally fast on security.

A separate Gravitee report found that 88% of organisations had confirmed or suspected an AI agent security or data privacy incident in the last twelve months, while 48% of AI agents in production are running with no security monitoring at all. Eighty-two percent of executives feel confident their existing policies protect them from unauthorised agent actions, despite the incident data showing otherwise.

That gap between executive confidence and actual exposure is where breaches happen.

What This Means for Business

If you are deploying AI agents in your business, whether for customer service, internal operations, data analysis, or any other function, the security posture you designed for traditional software needs to be revisited.

A few practical implications:

Your AI agents are attack surfaces. Every AI agent that can access your data, your systems, or your customers is a potential entry point. Attackers are actively looking for prompt-injection vulnerabilities, credential extraction opportunities, and ways to manipulate agents into taking unauthorised actions.

Speed of response is no longer enough. If exploits are being generated in hours rather than days, reactive patching is not a sufficient strategy. You need monitoring that catches anomalous agent behaviour in real time, not retrospective analysis after the breach.

Identity verification needs to evolve. If you rely on voice or video to verify customers, employees, or vendors, those checks are increasingly unreliable without additional layers. Multi-factor approaches that do not rely solely on biometrics are becoming essential.

Governance is not optional. The organisations that get through 2026 without a serious AI security incident are the ones that built governance into their AI deployments from the start, not the ones that bolted it on afterward.

The Check Point report is a useful reality check. AI is transforming what businesses can do. The same capabilities are being used to transform what attackers can do. The organisations that acknowledge this early and build accordingly are in the best position.

For businesses still in the planning stages of AI deployment, now is the time to get the security architecture right before the agents go live. For businesses already running AI in production, the audit of what you actually have running, and how it is monitored, should happen this week, not this quarter.

---

Enterprise DNA helps businesses deploy AI agents that are built for enterprise use, with governance and security built in from the ground up. If you are evaluating how to scale AI in your organisation without creating new exposure, start with a discovery conversation.