Summer’s hacking contest did not feel like a drill; it felt like an alarm bell.
What happened and why it matters
Chinese AI startup Z.ai released open-weight models that tore through attack surfaces with frightening efficiency. A local firm, Tenzai, used those models to win a three-month HackerOne contest by finding vulnerabilities in live networks belonging to government agencies, banks, hotels, airlines and more. The takeaway is brutal and simple: open-weight AI is a force multiplier — for both attackers and defenders.
That sentence should be read twice. Open-weight systems mean fewer guardrails, fewer restrictions, and far greater freedom to probe, synthesise, and adapt. These models do not merely speed up reconnaissance; they change the cadence of compromise. What used to take days can now be done in hours. What used to require a seasoned team can now be amplified by an AI assistant that suggests novel payloads, enumerates overlooked entry points, and composes tailored exploits.
Lessons for Singapore SMEs
SMEs in Singapore already operate in a tightly regulated and highly connected environment. That connectivity is an advantage for growth and a liability for exposure. The HackerOne outcome is a warning: attackers will adapt faster when powerful tools are readily available. But defenders can adapt too — faster, smarter, and with better economies of scale.
- Speed matters. Patch windows must shrink. Manual workflows that used to be acceptable are now a liability. Automate vulnerability scanning, triage, and patch deployment where possible.
- Adopt AI defensively. Use open models to run red-team exercises at scale, to prioritise risks, and to generate detection rules. The same capabilities that accelerate attacks can accelerate defence when harnessed correctly.
- Govern the tools. Not every model should be adopted blindly. Maintain an inventory of AI systems in use, define acceptable use policies, and require logging and explainability where feasible.
- Collaborate. Share indicators with peers via trusted channels. SMEs benefit when incidents are not siloed but treated as collective threat intelligence.
Anecdote from the field
One late-night incident remains vivid: a small hospitality client reported strange outbound traffic. A quick AI-assisted review revealed an automated reconnaissance script that had been fine-tuned to the hotel’s booking interface — it was probing for a specific parameter format used only in that property management system. Humans would have eventually found it, but the model found it within an hour. Heart racing, a weekend patch rolled out and a misconfiguration closed. Relief was intense. Relief was also humbling. The lesson: speed and visibility save reputations, but only when teams are prepared to move as fast as the new threat vectors.
Balancing the risk/reward equation
Experts are clear: more cyberattacks are coming. Accessible, powerful AI lowers the barrier to entry for attackers and enables them to scale. That is scary. It is also honest. The same AI landscape enables defenders to scan broader, reason faster, and predict with greater nuance. Balance will not materialise by itself. It will be engineered.
Expect a tug-of-war: offensive playbooks will proliferate in public forums and closed channels alike. Defensive tactics will improve, too, because defenders have incentives beyond disruption — reputational risk, regulatory compliance, customer trust. Over time, defensive tooling will become more integrated, AI-driven detection will mature, and playbooks will be shared across industries. That trajectory does not eliminate risk; it redistributes it toward those who move slower.
Practical steps that matter now
- Prioritise assets — not all systems deserve the same level of scrutiny. Map business-critical services and focus defensive investment there.
- Run frequent, AI-enhanced red teams — these are not academic exercises. They reveal asymmetric weaknesses and sharpen incident response.
- Harden identity and access — multi-factor authentication, least privilege, and session monitoring remain the most effective dampeners against automated attacks.
- Invest in logging and observability — AI finds patterns; humans act on signals. If telemetry is poor, detection will fail regardless of how advanced the models are.
- Educate staff with urgency — phishing will still be the easiest route for many attacks. Continuous training plus campaign simulation reduces human risk drastically.
Final stance
The HackerOne contest proved what many feared and what few were willing to admit out loud: openness in AI models accelerates capability transfer, and that transfer is neutral. It benefits whoever controls the playbook. Singapore SMEs must accept that neutrality and decide proactively which side of the ledger they want to inhabit. The choice is not between fear and denial. The real choice is between adapting aggressively or paying dearly for delay.
Act with urgency. Build with intent. Collaborate without hubris. Use the tools that are reshaping the battlefield to protect what matters most — customers, continuity, and trust. The next three months will be decisive. The next three years will be formative. Preparation now is not optional; it is strategic security.