In June, the Five Eyes intelligence alliance released an uncommon coordinated warning. Frontier AI models, the alliance cautioned, are "anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities." This assessment reflects a reality already evident in the field: attackers have demonstrated they can circumvent existing safeguards, and the technical barrier for less-skilled adversaries to launch increasingly complex attacks continues to erode.
The implications are stark for traditional security operations. Passive defense—monitoring and responding after threats emerge—no longer suffices against opponents equipped with AI-driven reconnaissance and attack generation. The challenge demands a fundamental shift in how security organizations approach their mission.
In this environment, defending attack surfaces won't cut it. We need to go on the offense.
The best defense is a good offense
Security professionals instinctively view the emergence of powerful AI models as a threat. Yet concern alone provides no protection. The reality is that most defensive security tools originated as offensive techniques, later repurposed for protection. Defenders possess equal access to advanced technology and can apply the same logic: if attackers deploy AI to map organizational assets and identify vulnerabilities, defenders should use identical methods to survey their own environments first.
This proactive stance emerges naturally from certain professional backgrounds rather than from deliberate methodology alone.
Different paths, different approaches
Chief Information Security Officers arrive at their positions through varied career trajectories. Some ascended from governance, risk, and compliance roles already managing organizational risk. Others came from IT operations, inheriting security responsibilities alongside broader technology management. Still others built expertise in security architecture, gaining deep knowledge of system design. An increasing cohort emerged from engineering backgrounds. Each pathway cultivates distinct perspectives on security challenges.
Engineering backgrounds tend to produce an attacker mindset, one that proactively searches for holes across the entire program the way an adversary would.
Compliance and IT-rooted leaders typically adopt a defensive posture: identify problems through logs, alerts, or incident response. Engineering-trained security leaders, particularly those from software companies, naturally gravitate toward an attacker's perspective—systematically hunting for weaknesses across the entire security program as an adversary would. Current circumstances demand the latter approach.
How to think like an attacker
Organizations with an attacker-oriented security culture construct scalable, automated solutions that deliver business outcomes while continuously uncovering gaps in programs, threat landscapes, infrastructure, and asset inventories. This requires genuine inquisitiveness: understanding system configurations, their rationale, and potential failure modes. Security research fundamentally rests on curiosity about how systems function, their intended behavior, and methods to produce unintended results.
Restructuring security functions around engineering principles and outcome-based metrics drives this transformation. Automated agents, guided by specific business objectives, identify risks, prioritize them, execute remediation, and generate reports. Application security teams shift toward engineering enablement, deploying agents that manage end-to-end tasks throughout the software development lifecycle. Governance and risk functions evolve into trust and resilience operations, continuously validating organizational security posture.
Building the right AI advantage
Two practical principles enable this transition. First, maintain model neutrality. Avoid locking workflows to specific models or vendors, since today's leading solutions may become obsolete within months. Deployment neutrality matters equally—security teams should retain options to operate models in air-gapped or self-hosted environments when data residency or intellectual property protection require it, not merely the ability to switch vendors.
Second, enforce strict scope boundaries. Assign agents narrow, well-defined tasks. Broad, vague problems cause agents to perform adequately initially, then deteriorate as they generate outputs they believe satisfy the prompt. Specific assignments, paired with high-quality contextual information, yield consistent, accurate results. This approach orchestrates intelligent automation without sacrificing precision.
Waiting for an incident or a vendor patch is a luxury we can no longer afford.
Organizations not yet prepared for agent-based restructuring need not delay action. For programs earlier in this transformation, the approach remains straightforward: examine each security function, define the specific outcome required, and work backward using existing tools rather than assuming an entirely new technology stack is necessary.
It's time to rethink things
The principal danger now is paralysis through overanalysis. The threshold attackers must overcome continues declining while the threshold defenders must clear rises. Deferring action until incidents occur or vendors release patches represents a luxury security programs can no longer sustain. Adapting security operations to match this moment is essential to avoid becoming casualties of adversaries' technological evolution.