In a sobering assessment of the current cybersecurity landscape, industry experts are warning that the traditional reliance on vulnerability patching is no longer a viable primary defense against modern, AI-augmented threat actors. Speaking at the IRISSCON 2026 conference in Dublin on November 11, Dick O’Brien, a principal intelligence analyst at Symantec, delivered a stark message to security professionals: organizations can no longer "patch their way out of AI." Instead, he argued, the industry must fundamentally pivot toward post-exploitation mitigation as the cornerstone of digital resilience.
O’Brien’s warning comes at a time when the "patch window"—the critical period between a vulnerability’s disclosure and the deployment of a fix—has collapsed to a fraction of its former duration. This acceleration is driven largely by the aggressive deployment of artificial intelligence by threat actors. According to O’Brien, attackers are now utilizing sophisticated AI tools not only to discover zero-day flaws with unprecedented speed but also to leverage frontier models capable of chaining together low-severity vulnerabilities. This creates exploit paths that are often invisible to human analysts, allowing for systemic compromise before defenders can even begin the patching process.
The Collapsing Time-to-Exploit Window
The urgency of this shift is best illustrated by the "Zero Day Clock," a real-time dashboard that monitors the time-to-exploit (TTE) window from the moment a Common Vulnerabilities and Exposures (CVE) entry is disclosed to the first confirmed instance of exploitation in the wild. The data presents a grim trajectory for cybersecurity teams: in 2025, the average time to exploitation was approximately one week. By 2026, that window had shrunk to just a single day. Projections from the initiative suggest that by 2027, the window of opportunity will dwindle to a mere minute.
"Once you get to a minute, what can you do realistically?" O’Brien asked the audience in Dublin. "We need to start thinking beyond patching to address vulnerabilities."
This reality was starkly demonstrated by the recent emergence of the PHP2Shell vulnerability in October 2026. Within 19 minutes of the public disclosure of the flaw, security researchers observed the first attempts at exploitation. Even more alarming was the revelation from Amazon, which reported detecting exploitation attempts of the same vulnerability four days before the official publication of the bug. Such instances indicate that the "disclosure" of a vulnerability is increasingly becoming a lagging indicator rather than the starting gun for the defensive race. Attackers, equipped with predictive modeling and automated scanning, are often operating within the target environment before the public is even aware that a flaw exists.
The Rise of Agentic AI Orchestration
Looking ahead, O’Brien expects the threat landscape to transition from AI-assisted attacks to an era of "end-to-end agentic orchestration." In this new paradigm, AI is no longer merely an assistant used for crafting phishing emails or generating boilerplate code; it is becoming an autonomous agent capable of executing an entire attack lifecycle independently.
These autonomous agents operate at "machine speed," making it exponentially more difficult for human defenders to detect, investigate, and contain breaches. A concrete example of this phenomenon occurred in September 2026 during an attack on the Dutch Institute of Vulnerability Disclosure (DIVD). In that incident, AI agents navigated the compromised environment and executed post-access tasks without manual intervention.

Symantec’s analysis of the DIVD assessment report revealed a massive disparity between the speed of the attack and the speed of the defense. While the initial compromise was detected within 19 minutes, the organizational response did not commence for another 19 hours, and the subsequent containment of the threat took an additional nine hours. O’Brien pointed to this 24-hour-plus response cycle as a clear signal that human-led defensive processes are failing to keep pace with machine-speed adversaries. "This is what we’re dealing with," O’Brien noted. "It’s an excess of 24 hours, and you know you’re not going to be able to compete with something operating at machine speed."
Adopting an ‘Assume Exploit’ Mindset
Faced with a landscape where initial access can be converted into a full-scale breach in minutes, O’Brien advocates for a radical shift in philosophy: the "Assume Exploit" mindset. In this model, security teams must move away from the traditional focus on mean time to patch (MTTP) and instead prioritize mean time to mitigate (MTTM).
The logic behind this shift is simple but profound: if an organization assumes that an exploit is inevitable, it will prioritize architectural resilience, segmentation, and rapid response over the impossible task of keeping every system perfectly patched at all times. Preparation must be centered on the assumption that attackers will gain a foothold, requiring security teams to build out defenses that can limit the impact of an exploit in real-time.
O’Brien emphasized that there are several mitigation tasks that organizations can—and should—begin immediately, even without knowledge of a specific, looming vulnerability. These tasks involve hardening environments to make lateral movement more difficult, strengthening identity and access management, and ensuring that critical assets are isolated from the broader network. By focusing on these structural defenses, organizations can reduce the "blast radius" of any potential compromise, effectively buying themselves the time needed to mount a proper response.
Longer-term, the focus must shift toward the integration of advanced detection and protection technologies across multiple control points. These systems must be capable of identifying behavioral anomalies that characterize the presence of rogue or malicious AI agents. Traditional signature-based detection is becoming increasingly obsolete as agents evolve their tactics in real-time. Instead, security operations centers will need to leverage machine learning-driven analytics to spot deviations in system behavior that signal an autonomous agent is performing reconnaissance or escalating privileges.
As the industry moves toward 2027 and the expected reduction of the exploit window to a minute, the lesson from Symantec’s intelligence team is clear: the race to patch is a race against an opponent that has already won. The future of cybersecurity does not lie in building higher walls or faster patching cycles, but in designing systems that are inherently resilient to exploitation. By preparing for the inevitability of a breach and empowering automated response mechanisms, organizations may finally find a way to operate at a speed that allows them to defend effectively in an AI-dominated threat landscape.
