How companies can stay ahead of AI security risk

By Abhishek Kumar Singh, Head of Security Engineering, Singapore, Check Point Software Technologies

Everyone wants to move fast with artificial intelligence (AI). Yet, as organisations race to roll out new AI capabilities, a stark reality is emerging: AI doesn't just introduce new security risks, it acts as an amplifier for all the architectural flaws we’ve already been sweeping under the rug.

Industry forecasts suggest that AI-related security incidents could drive up to a quarter of all enterprise breaches by 2028. For IT and security leaders, rushing into AI adoption without cleaning up foundational hygiene is an open invitation for disaster.

The ground shift

"The latest Annual AI Security Report 2026 published by Check Point Research documents a decisive shift over the past twelve months; AI has moved from assisting attackers to operating attacks. Where AI once helped criminals prepare, it now runs live intrusions with minimal human direction," says Abhishek Kumar Singh, Head of Security Engineering, Singapore, Check Point Software Technologies.

Grounded in real incidents, telemetry, and original case studies, the report highlights how AI participates directly at every stage of the attack chain, compressing the time defenders have to respond and opening new attack surfaces as enterprise adoption outpaces governance controls.

Key findings from the report:

AI is now operating attacks, not just enabling them

Researchers documented intrusions where AI ran exploitation workflows autonomously. In one breach of nine Mexican government agencies, a single operator combined Claude Code (for network exploration) with GPT-4.1 (which generated 5,317 AI-executed commands across 34 attack sessions) to analyse stolen data and task follow-on activity.

The vulnerability window has collapsed from days to hours

AI can turn a fresh vulnerability disclosure into a working exploit within hours, prompting authorities to shorten mandated remediation timelines to as little as 12 hours for critical Internet-facing systems.

Prompt-injection payloads surged by roughly fivefold between March and May 2026

The sharp increase in large malicious payloads is consistent with indirect prompt injection becoming a routine attack path and operational enterprise risk rather than a theoretical one, as AI itself becomes an attack surface.

Identity can no longer be trusted as standalone security control

With voice, face, and real-time video easily synthesised, with highly-trained reviewers correctly spotting AI-generated faces only 41% of the time, organisations must move beyond visual verification toward stronger identity assurance and out-of-band verification methods.

High-risk enterprise AI prompts have doubled over the year

High-risk enterprise AI prompts climbed from roughly one in 50 interactions to one in 25. While the average organisation runs 10 AI apps a month, many unapproved, between 87% and 93% experience at least one high-risk interaction monthly.

Most enterprise data exposure comes from ordinary, approved use, not from attacks, as employees share more context than they realise to get a useful answer.

Modern AI tools don’t live in a vacuum. They hook directly into our collaboration platforms, internal knowledge bases, cloud repositories, application programming interfaces (APIs), and business applications.

Every single one of those integrations opens up a new pathway. The problem is that AI is exceptionally good at finding, summarising, and redistributing information at a massive scale. If you already have structural issues, like sloppy data segmentation, sprawling service accounts, or over-permissive access rights, AI weaponises those flaws instantly.

Imagine a legacy service account with overly broad permissions connected to an internal database. Historically, that might have stayed a quiet, low-priority configuration oversight. Give an AI assistant rapid, query-level access to that same database, and that minor blind spot turns into a wide-scale data exposure event overnight.

Before you give your entire workforce the keys to advanced AI tools, security teams need to step back and validate that the fundamentals actually work. You don't need to stall innovation, but you do need to anchor it to a secure foundation.

Prioritise these core areas first:

- Tighten identity controls: Audit everything with access; human users, admins, third-party service accounts, and the AI integrations themselves.

- Enforce least privilege: Make sure your AI applications can only touch the specific data and functions required for an authorised workflow, nothing more.

- Map and classify your data: Don't plug external large language model (LLM) pipelines into repositories until you genuinely know what sensitive data lives inside them and who can see it.

- Upgrade your telemetry: If you can't monitor, log, and alert on your AI-connected APIs and cloud services in real time, you're flying blind.

- Manage configuration drift: Environments change constantly. Your security guardrails need to adapt just as fast as your developers and users deploy new tools.

In modern cloud and AI ecosystems, static security is dead. A point-in-time audit gives you a nice snapshot of your risk posture on a Tuesday afternoon, whereas by Wednesday, a developer has spun up a new integration, and configuration drift has already set in.

Building genuine resilience requires continuous validation. Security teams need ongoing visibility, automated control checks, and fast remediation workflows that scale alongside AI adoption rather than lagging behind it.

The winners in the AI era won't necessarily be the ones who cross the finish line first. Instead, they'll be the ones who manage to scale securely. By locking down identity management, enforcing least-privilege principles, and shifting to continuous validation early on, you can stop playing catch-up and start driving AI transformation with real confidence.