If your team has quietly let AI tools into its daily workflow, a new report from Check Point Research argues the security risk has changed shape. Attackers are no longer only using AI to write sharper phishing emails. They are handing it the keyboard.
Editor
Kai T chevron_right
Table of Contents
What the Check Point AI Security Report 2026 found
Check Point Software Technologies published its Annual AI Security Report 2026 on 14 July, drawing on real incidents, telemetry and case studies gathered over the past twelve months. The central finding is blunt: AI has moved from assisting attackers to operating attacks, running live intrusions with minimal human direction between steps.
The clearest example in the report is a breach of nine Mexican government agencies. According to industry reports cited by Check Point, a single operator paired two commercial AI tools, using one to break in and explore networks and the other to analyse stolen data and task follow-on activity. Together they generated 5,317 AI-executed commands across 34 attack sessions.
The patch window has collapsed to hours
Check Point says AI can now turn a fresh vulnerability disclosure into a working exploit within hours, not days. Government authorities have responded by shortening mandated remediation timelines to as little as 12 hours for the most critical internet-facing systems, which is a brutal standard for any IT team that still patches on a monthly cycle.
The AI tooling itself is also becoming a target. Detections of long, malicious prompt-injection payloads rose roughly fivefold between March and May 2026, which the researchers read as a sign that indirect prompt injection has become a routine attack path rather than a lab curiosity.
Identity checks are getting harder to trust
Voice, face, documents and real-time video can now be synthesised convincingly enough that highly trained reviewers correctly identified only around 41 percent of AI-generated faces. Check Point argues that visual verification alone is finished as a control, and that organisations need multi-factor authentication and out-of-band verification instead of trusting what they see on a video call.
The bigger leak is the everyday one
The number that should worry most businesses has nothing to do with hackers. High-risk enterprise AI prompts doubled over the year, from roughly one in every 50 interactions to one in every 25. The average organisation now runs ten AI applications a month, many of them without formal approval, and between 87 and 93 percent see at least one high-risk AI interaction every month. Most data exposure in the report came from ordinary, approved use, from employees pasting more context than they realised into a chatbot to get a useful answer.
Lotem Finkelstein, Vice President of Check Point Research, put it this way: "The expertise barrier that separated capable attackers from the rest is disappearing, and defenders can no longer assume a human is setting the pace on the other side."
What it means for you
Check Point frames its answer around three ideas: securing the AI systems a company now depends on, defending at machine speed with AI, and governing how staff actually use AI. That is a vendor pitching its own product line, so treat the prescriptions accordingly. The underlying observation is harder to dismiss. If a mid-sized Malaysian company has ten unapproved AI tools floating around its teams, the exposure is already inside the building, and no amount of perimeter security addresses it.
The practical takeaway is unglamorous: know which AI tools your people are using, patch faster than you are comfortable with, and stop treating a familiar face on a video call as proof of identity.
Cover photo by Compagnons on Unsplash. Body photo by Taylor Vick on Unsplash.