AI-powered attackers don’t wait: What’s next for Singapore’s schools?

In my previous discussions on the state of cybersecurity in Singapore’s schools, I explored the “GenAI paradox” — how the very tools driving classroom innovation are simultaneously supercharging the capabilities of threat actors.

I argued that the education sector had arrived at a point where institutions’ legacy perimeter defenses no longer kept pace with modern cybercriminals.

We’re now past that point.

Cybersecurity in education just became harder

The security landscape has shifted from standard generative tools to autonomous frontier AI models, exemplified by tools like Anthropic’s Claude Mythos.

Whereas human adversaries took weeks to research an institution and find an unpatched vulnerability to exploit, these models have greatly compressed this entire attack lifecycle. According to Fortinet’s 2026 Global Threat Landscape Report, AI-fueled cybercrime has reduced the average time to exploit critical vulnerabilities to as little as 24 to 48 hours.

As Singapore’s Coordinating Minister for National Security, K Shanmugam, and the Cyber Security Agency of Singapore (CSA) have repeatedly emphasized, cyber resilience has become a board-level priority.

The CSA has warned that AI can compress vulnerability research timelines from months to hours, making externally exposed assets the most urgent area for risk reduction.

Educational institutions already balance strict public-sector budgets with complex, open-campus networks. Now they must deploy cost-effective and highly agile defenses to keep up with automated attacks.

• ALSO READ → Cyber resilience in education: Why the sector is at a critical inflection point

The new reality: Automated reconnaissance and hyperspeed exploits

Traditional cybersecurity frameworks operate on the assumption of human latency. If a new vulnerability is discovered, an IT team typically has a window of a few days or weeks to test and deploy a patch before threat actors build a weaponized exploit.

That assumption is breaking down. Microsoft Threat Intelligence reports that threat actors are moving beyond experimentation and increasingly integrating AI into reconnaissance, malware development, social engineering, post-compromise activities and agentic attack workflows.

Now, when a malicious autonomous AI agent finds a flaw, it can autonomously write a targeted exploit and execute it within hours of discovery.

That’s where frontier AI models are proving invaluable. Autonomous models like Mythos excel at scanning vast ranges of public-facing infrastructure simultaneously. They find open ports, analyze the target context, synthesize code on the fly and systematically probe for weaknesses in:

• Internet-facing assets: Public web portals, legacy content management systems and unpatched virtual private network gateways
• Exposed development environments: Staging servers, forgotten sandboxes and shadow-IT cloud instances created during short-term academic research projects
• Third-party application programming interface (API) integrations: Connected educational technologies and vendor platforms that bypass standard internal security baselines

If an institution’s public-facing attack surface is visible to the internet, a frontier AI model can often map much of it before a security operations center even detects reconnaissance activity.

The CSA framework: Hardening the perimeter safely and affordably

To defend against AI-driven attack speed, educational institutions don’t need to invest millions in opaque, enterprise-grade AI defense tools. Instead, they need to focus on what the CSA calls high-priority, foundational cyberhygiene.

When organizations neutralize an attacker’s ability to easily gain a foothold, they render the autonomous model’s capabilities ineffective at the outer perimeter. In academic environments, this defense blueprint must be simple to maintain and highly cost-effective:

1. Attack-surface reduction in real time

If a frontier AI model cannot see an asset, it cannot exploit it. Most organizations have major blind spots and limited visibility of shadow IT and internet-exposed assets. Traditional tools often struggle to map the complete external attack surface.

Institutions must therefore prioritize continuous discovery to maintain complete visibility of their external digital footprint. Every internet-facing web server, API endpoint and cloud database needs to be identified and cataloged.

In addition, any nonessential testing or staging environments should be immediately disconnected from the public internet or placed behind a strict corporate firewall.

According to Qualys, organizations commonly discover an additional 30%–40% of previously unknown internet-facing assets when implementing external attack-surface management, highlighting the significant visibility gap that exists in many environments.

2. Shorter patching windows for critical vulnerabilities

Because frontier models dramatically shrink the timeline between vulnerability disclosure and active exploit, institutions must establish automated, high-priority patch pipelines.

The goal is not to remediate every vulnerability immediately but to focus resources on the vulnerabilities that are most likely to be exploited. Prioritizing internet-facing assets and vulnerabilities with confirmed exploitation activity enables organizations to reduce risk faster and more efficiently.

The US Cybersecurity and Infrastructure Security Agency (CISA) has shown that prioritizing known exploited vulnerabilities can significantly reduce long-lived exposures and speed up remediation, while focusing on internet-facing assets helps address the attack paths most likely to be targeted by adversaries.

3. Strict perimeter isolation and zero trust controls

Autonomous tools excel at exploiting single points of failure, such as compromised administrative credentials. Enforcing strict, phishing-resistant multifactor authentication for every single access point is nonnegotiable.

Furthermore, internal networks must be segmented. If an AI agent manages to penetrate a public-facing system, strong internal network microsegmentation ensures the blast radius is fully contained, preventing the exploit from moving laterally into critical student databases or financial management systems.

Multifactor authentication (MFA) can prevent more than 99% of account compromise attacks. Microsoft research found that MFA reduces the risk of account compromise by 99.22%, while CISA states that users who enable MFA are 99% less likely to be successfully compromised.

NTT DATA helps protect educational institutions

Singapore’s educational institutions operate under distinct operational and budgetary realities. Schools cannot sacrifice the open, collaborative spirit essential to learning, yet they must remain steadfast against increasingly sophisticated threats.

True resilience in the era of frontier AI comes down to executing the fundamentals quickly and consistently. By working with expert service providers like NTT DATA to prioritize attack-surface visibility, speed up critical patch management and embed robust perimeter controls, institutions can navigate this threat landscape without overextending their operational budgets.

The window of exposure has shrunk to nearly zero. The time to strengthen the perimeter is before the next vulnerability is discovered — not after it’s being exploited.

• ALSO READ → Trust at the speed of AI: Building a cybersecurity strategy for AI