AI-driven cyberattacks concept with glowing AI letters on a circuit board
Digital Tools & Enterprise TechWork & Industry

AI-Driven Cyberattacks: How Agentic AI Is Reshaping the Threat Landscape

URL copied
Share URL copied

AI-driven cyberattacks are moving beyond smarter phishing. Malis Selamat, Managing Director, ASEAN at Sophos, explains how agentic AI is shrinking the time defenders have to respond.

Most talk about AI and cybersecurity has focused on better phishing emails or cloned voices for fraud calls. That is real, but it understates what is happening now.

What’s different about agentic AI is autonomy. An attacker can hand an AI agent a goal rather than a script. The agent then scans for exposed services, adapts when a defence blocks it, and chains together several stages of an intrusion with little human input.

AI-Driven Cyberattacks Now Run Like a Pipeline

A threat intelligence disclosure in June 2026 made this concrete. Researchers uncovered a working threat-actor lab built to develop malware that defeats endpoint detection tools.

An AI “manager” directed several agents through code generation, testing and revision. It also ran each build against live security products before shipping it. This wasn’t theoretical red-teaming. Instead, it was a pipeline that industrialised what used to be slow, manual work.

AI-driven cyberattacks concept with glowing AI letters on a circuit board
Agentic AI lets attackers hand autonomous agents a goal rather than a script.

Identity Attacks and a Three-Day Window

The bigger story sits in the latest Sophos Active Adversary Report, drawn from 661 incident response and MDR cases. Median attacker dwell time has fallen to just three days. Meanwhile, stolen credentials, weak MFA and poorly protected accounts now beat software exploitation as the main way in.

Sophos’ State of Identity Security 2026 report agrees. It found that 71% of organisations suffered at least one identity-related breach in the past year. Also, 67% of surveyed ransomware victims traced their incident to an identity attack.

In other words, AI isn’t inventing new ways through the door. Instead, it collapses the time between getting in and doing damage. For defenders on human-paced response cycles, that is the more dangerous problem.

Malaysia’s numbers point the same way. CyberSecurity Malaysia’s Cyber999 centre logged a 29% jump in reported data breach incidents in the first quarter of 2025 alone.

Why This Is a Governance Problem, Not Just a Tooling One

Malaysia’s regulators are asking similar questions. The Cyber Security Act 2024 and Bank Negara Malaysia’s (BNM) revised RMiT policy have raised the bar on incident reporting, risk assessment and controls. Meanwhile, the Personal Data Protection (Amendment) Act 2024 has made breach notification mandatory since June 2025, with higher penalties.

The rules are still evolving. In July, Parliament passed the Cybercrimes Bill 2026 to replace the Computer Crimes Act 1997, including to tackle the misuse of AI. The Digital Ministry has also completed a draft AI Governance Bill, due in Parliament by early 2027 at the latest. Together, these moves show that AI and cybersecurity are now governance and accountability issues, not just decisions for security teams.

Frameworks alone don’t close the human gap, though. In January 2026, BNM fined Bank Kerjasama Rakyat Malaysia Bhd RM1 million after an external breach. It cited weak cybersecurity controls and incident response, not a missing policy.

Most governance frameworks assume human-paced threats. However, AI-driven cyberattacks move, adapt and escalate at machine speed. So organisations must rethink not just their tools, but how fast decisions and response can happen.

How to Defend Against AI-Driven Cyberattacks

None of this calls for panic. AI-driven cyberattacks don’t change the fundamentals; they make them matter more. Four priorities stand out:

  • Identity first. Identity compromise is now the leading entry point. So phishing-resistant MFA and tight control of human and machine identities belong at the top of any 2026 security roadmap.
  • Assume compressed timelines. With a three-day dwell time, detection, containment and response must run in parallel, not in sequence. That usually means 24/7 managed detection and response rather than a business-hours security team.
  • Match machine speed with machine speed. An AI-assisted pipeline can test dozens of evasion variants in an afternoon. Therefore, detection must track attacker techniques and behaviours, not just known signatures, because the payload may be new every time.
  • Close the human gap. Tools without trained people and forensic readiness leave the weakness that BNM’s fine put a price on. As a result, tabletop exercises and response drills matter more than ever.

Agentic AI hasn’t replaced the attacker’s human intent, and human error still drives many incidents. What it has done is remove the friction that once gave defenders breathing room. That room now needs to be re-engineered into enterprise defence, through stronger identity controls, always-on monitoring and behaviour-based detection, before it disappears for good.

Read more: how a widening cybersecurity skills gap is catching out security teams across APAC.

Written by
Dewi Lestari

Dewi Lestari covers Living Asia for RiseAsia — how life in the region's cities is changing, from culture, healthcare and cost of living to property and housing, infrastructure and mobility, urban development and consumer technology.