From the Frontlines: Why SecOps and Incident Response Need Prevention in the Cloud
In the incidents I handled, the adversary rarely needed a zero-day. They only needed one setting to be wrong. In the cloud, that setting can expose an entire data store, sometimes for just a few minutes, and attackers now move faster than detection can: the average breakout takes 29 minutes, the fastest 27 seconds. Here is what I saw from the frontlines, and why SecOps and incident response teams should start with prevention.

Every organization can expect to deal with a security incident, and in my experience, most incidents share a common denominator: a misconfiguration somewhere along the attack path. It may be a permission that is broader than it needs to be, a security control that was disabled, or a resource that should never have been reachable from the internet. As a previous incident handler, I worked through incidents where the adversary did not need a zero-day exploit. The adversary only needed one setting to be wrong.
The same pattern now exists in the cloud, with one major difference. Adversaries are using AI to find and exploit misconfigurations at machine speed, and security operations centers (SOCs) cannot keep up using detection alone. In this post, I will walk through a few recent cloud incidents and explain why security operations (SecOps) and incident response (IR) teams need a preventive approach in the cloud.
Cloud misconfigurations at scale
In the cloud, a single misconfiguration can expose an entire data store to the internet. The following examples are publicly disclosed.
Indian bank transfer records
In late August 2025, UpGuard researchers discovered a publicly accessible Amazon S3 bucket containing approximately 273,000 PDF documents related to bank transfers processed through India's National Automated Clearing House (NACH). The documents included account numbers, transaction amounts, and contact details. The bucket was secured in early September, and after the disclosure on September 25, the Indian fintech Nupay confirmed it had addressed a configuration gap in an S3 bucket. Nupay did not say how long the bucket was publicly accessible.
Toyota
In May 2023, Toyota disclosed that a cloud environment managed by its subsidiary, Toyota Connected Corporation, had been set to public instead of private due to human error. The misconfiguration exposed vehicle location and time data, in-vehicle GPS navigation terminal ID numbers, and chassis numbers for approximately 2.15 million customers in Japan. The data was accessible without a password from November 6, 2013, to April 17, 2023. Toyota stated it had no reports of the data being misused.
From an IR perspective, the length of the exposure is the most concerning part of this incident. One cloud setting remained wrong for nearly a decade, and in response, Toyota said it would introduce a system to monitor its cloud settings and audit its cloud environment. Monitoring helps find the next mistake, but it still depends on someone reviewing the finding and remediating it. A preventive guardrail that denies making the data store public would have blocked the misconfiguration on the day it was introduced. This is the type of control Aryon enforces at the cloud control plane, and my colleagues covered this incident in more detail in Ten Years Is a Long Time to Be Wrong.
Cloud ShutterGap research
Some cloud exposures last only a few minutes, which may not sound like enough time for an adversary to act. Research from my team at Aryon shows otherwise. In July 2026, Aryon published the Cloud ShutterGap research, which monitored AWS resources that support public sharing, such as RDS and DocumentDB snapshots, Amazon Machine Images (AMIs), and Systems Manager (SSM) documents. The research found that 20% of publicly shared RDS snapshots appeared and disappeared in under two minutes, and 99% of the RDS and DocumentDB snapshots that were later deleted were removed within 30 minutes of creation.
For example, an adversary can continuously enumerate public RDS snapshots, copy a newly shared snapshot into their own AWS account within seconds, and restore the database later. The exposure only needs to last long enough for the copy to start. In a sample of 24 public snapshots, the team found AWS account IDs in every snapshot, along with email addresses, potential secrets, and patterns associated with private keys and financial data. It is important to note this is not an AWS vulnerability. These are customer misconfigurations, and AWS documentation warns against storing sensitive data in publicly shared resources.
The challenge with a detection-only approach
Cloud security posture management (CSPM) and cloud-native application protection platform (CNAPP) tools play an important role in providing visibility. However, a detection-only approach has four problems that I see SecOps and IR teams run into again and again.
It assumes you have time to remediate
CSPM and CNAPP tools typically rely on periodic scans, often once every 24 hours. If a resource is exposed and removed between scans, the tool never generates a finding. Even when a finding is generated, the alert still needs to be triaged, assigned, and remediated, and the adversary may have already copied the data. The window is often measured in minutes. One of my colleagues at Aryon tracked this at a previous company: a virtual machine with RDP open to the internet and a weak password was taken over by automated bots within minutes. According to the CrowdStrike 2026 Global Threat Report, the average eCrime breakout time fell to 29 minutes in 2025, the fastest observed breakout took 27 seconds, and operations by AI-enabled adversaries increased 89% year over year.
Detection without a standard is just noise
If you do not have a clear standard for what your cloud environment should look like, every deviation becomes an alert, and the findings that matter get lost in the noise. SOCs are already inundated with alerts, and adding more detection rules creates more alerts for analysts to review. Once you define a standard, the logical next step is to enforce it. Many teams defaulted to detection because the native preventive controls are difficult to deploy and manage at scale without breaking production. That is the gap Aryon was built to close.
It shifts the purpose of the security team
The cloud is always moving, with new services, new resource types, and new features released constantly. Security teams should be figuring out how to adopt these services securely and defining the standard for using them. Instead, a detection-only approach keeps them putting out fires from existing mistakes, leaving no time to get ahead of the next one. SOCs cannot hire their way out of this problem.
Remediation is expensive
Every software development methodology agrees on one point: fixing something after it is built costs more than getting it right the first time. Remediating a misconfigured production resource often requires a migration or a maintenance window, so remediation moves slower than new deployments. The result is a remediation backlog that grows faster than the team can work it down, an endless loop.
Organizations need a security in-depth strategy that provides prevention, detection, and response capabilities. Detection and response remain critical for the activity that gets through. However, prevention needs to be the first layer, stopping known-bad configurations before they exist so the SOC can focus on the alerts that matter.
Because of AI, organizations should now start with prevention. Historically, many security programs were built detection-first: collect the logs, write the correlation rules, and remediate findings as the team has capacity. That approach assumed the time between a misconfiguration and its exploitation would be measured in days or weeks. As adversaries use AI to automate reconnaissance and exploitation, that window is closing. Organizations should begin by enforcing preventive guardrails for known-bad configurations, and then build detection and response around the activity that prevention cannot stop.
Prevention and the incident response lifecycle
The incident response lifecycle from NIST SP 800-61 consists of preparation; detection and analysis; containment, eradication, and recovery; and post-incident activity. As a previous incident handler, I found the post-incident activity phase was the one most often rushed. The incident report is written, the lessons learned meeting takes place, and the recommendations are added to a backlog. Months later, the same type of misconfiguration appears in a different account.
One of the most important outcomes of an incident is ensuring it does not happen again. If the root cause is that a user was able to make a database snapshot public, the long-term remediation is a preventive control that blocks the action across the organization.
In AWS, this is typically a Service Control Policy (SCP), for example, denying rds:ModifyDBSnapshotAttribute or requiring storage encryption so snapshots cannot be shared publicly. Writing the policy is the easy part. The challenge is deploying it across hundreds of accounts without disrupting production workloads.
How Aryon helps SecOps and IR teams
SecOps and IR teams now have a way to turn incident findings into enforced guardrails. Aryon enforces policy at the cloud control plane, which blocks a misconfiguration before the resource is provisioned instead of reporting it after the fact. Aryon complements the CSPM or CNAPP you already use. If your CSPM is the intrusion detection system (IDS) for your cloud, Aryon is the intrusion prevention system (IPS).
- Close the loop on post-incident activity-the root cause identified during an incident becomes an enforced guardrail across your cloud environment instead of an item in the remediation backlog.
- Reduce repeat alerts-when a misconfiguration is prevented, it no longer generates the same alert in your CSPM and SIEM, which gives analysts more time to investigate threats that require a human decision.
- Deploy without disrupting the business-preventive controls are tuned and tested before enforcement so security and operations teams can implement guardrails without breaking production.
Conclusion
Throughout my career in SecOps and IR, misconfigurations have been a common denominator in the incidents I worked on. Adversaries leveraging AI have reduced the time between a misconfiguration and its exploitation from days to minutes, and in some cases seconds. A preventive approach in the cloud gives SecOps and IR teams a method to confidently close an incident, knowing the misconfiguration the adversary used cannot be reintroduced.
Frequently Asked Questions
Notes & Sources
- UpGuard, Unclaimed Property: How an Unknown Entity Exposed Indian Banking Information (Sept 2025)
- TechCrunch, Thousands of Indian bank transfer records found online (Sept 2025)
- SC Media, Over 2.1M Toyota Japan customers' data exposed in decade-long breach (May 2023)
- Aryon Security, Cloud ShutterGap: Millions of Cloud Resources Exposed (July 2026)
- CrowdStrike, 2026 Global Threat Report (Feb 2026)






.avif)