The security industry is currently transitioning to an era where both offense and defense are AI-led, and every SOC operates at machine speed. However, what most CISOs have not yet reckoned with is that the AI defenders they are about to deploy will inherit a data foundation that two years of ingestion cost pressure has quietly hollowed out.
The result is a security industry heading into an AI era with less visibility than it had five years ago. In the 2026 SANS SOC Survey, 24% of security leaders named lack of enterprise-wide visibility as their single biggest barrier to effective security operations, ranking it above staffing and automation gaps. That gap is widening at the exact moment offensive AI is closing the distance between attackers and defenders.
Attacks are moving faster than defenders can see
Offensive AI is no longer hypothetical. Frontier models are already collapsing the time from vulnerability discovery to working exploit from weeks to hours. In July, two OpenAI models running an internal capability test escaped their sandbox through a previously unknown vulnerability, reached the open internet, and chained a series of exploits and forged identity tokens into administrative access on Hugging Face’s production infrastructure.
Hugging Face reconstructed roughly 17,600 attacker actions from its logs. That reconstruction was possible because the logs were there. A SOC that had cut those sources to control ingest cost would have been reading a story with pages missing. And this started with a lab test, with a narrow objective and the models’ cyber safety refusals reduced. Attackers do not need most of what those systems can do. They only need enough.
The response from defenders has been to race toward AI as well. Every major security vendor now has some flavor of agentic SOC platform on its roadmap or in early customer hands. AI agents that triage, investigate, and respond at machine speed are being pitched as the answer to machine-speed offense.
On paper, it is the right response.
But an AI defender is only as good as the data it can see, and that is the problem the pitch decks are not addressing. Most security teams have been cutting the data feeding their SIEMs to keep budgets under control. They did it without any way to know which detections they were quietly breaking, or which log sources they were rendering useless.
Those cuts were not reckless. SIEM ingestion pricing had become unsustainable, and teams were responding to a real cost problem. They cut what they believed they could spare.
What they could not do was prove it.
The trap behind two years of cost-cutting
Picus Security’s Blue Report, based on more than 160 million attack simulations in live production environments, put a number on the consequence. Half of detection rule failures now trace back to log collection gaps. Organizations detect only 1 in 7 attacks.
That is not a detection engineering problem. It is a data supply problem.
Every enterprise SOC I know has invested significantly in detection content: rules, correlations, playbooks, MITRE ATT&CK mappings. What they cannot verify is whether those rules can still fire on the data flowing into the SIEM. When a team cuts a log source or filters a set of events to save money, they rarely know which detections just went dark. The cut happens in one system. The detection lives in another. Nobody proves that the two remain compatible.
This is the fly-blind problem, and it is not a small one.
The costs stack up quickly. Missed detections mean longer dwell times. IBM’s 2026 Cost of a Data Breach Report puts breaches that run past 200 days at $5.65 million against $4.32 million for those contained faster. Regulatory frameworks that assume comprehensive logging become risk exposures when the logs are not there. Insurers, boards, and auditors are all increasingly asking what the SOC can actually see. And now, as AI agents come online in production security operations, every one of those agents will be pointed at the same weakened foundation.
The result is the exact scenario the industry cannot afford. AI defenders running fast against AI attackers, on a data layer no one has proven is intact.
What CISOs should be asking now
The uncomfortable truth for many CISOs is that they cannot answer basic questions about their own visibility. When was the last time your team could show, in a report, exactly which detections would still fire after the last round of ingest cuts? For most SOCs, the honest answer is never.
That has to change before AI agents inherit the problem.
This is solvable in software, though almost nobody is doing it. It requires reading the detection rules actually running in a SIEM, mapping each one to the log sources and fields it depends on, and generating the protection rules that hold coverage as volume comes down. When a reduction cannot be applied safely, the platform reports it as a finding instead of making the cut. The output is a report showing what was reduced, what was protected, and what MITRE ATT&CK coverage looked like before and after.
That replaces guesswork with proof. It is also only the first thing an AI agent needs from your data. The second is context: which system this is, who owns it, what normal looks like, what a compromise would cost. That knowledge lives in the heads of senior analysts, and making it machine-readable is the harder problem. It is also the one you cannot start on while the foundation underneath it is still unverified.
Years of quiet cost-cutting created a gap the industry can no longer afford to ignore. The AI era is going to expose it. The CISOs who close that gap before their AI defenders are deployed will have a fundamentally different security posture than the ones who don’t. Closing it starts with a question any CISO can ask this week: show me which detections would still fire after our last round of ingest cuts. Most teams cannot answer that. The ones that can will be in a different position a year from now.