Archive for November 19, 2026

Gartner: 95% of orgs have had at least 1 AI agent incident

Posted in Commentary with tags on November 19, 2026 by itnerd

Gartner predicts that without strong governance, 40% of infrastructure and operations (I&O) organizations scaling AI will suffer a business-critical service disruption by 2028, due to survey findings just published this week. Its survey of 298 of I&O leaders from the U.S. and U.K. found:

  • 95% of organizations surveyed have faced at least one agent-related incident in the past 12 months.
  • There’s a massive operational readiness gap as 70% of I&O leaders believe their AI agents currently operate within safe boundaries.
  • 57% of I&O leaders said their current governance is weak.
  • 55% stated that existing IT risk tools are completely outdated
  • Only 15% have automatic circuit breakers in place to stop runaway agent errors or infinite logic loops.

Ryan McCurdy, VP, Liquibase:

“Gartner’s prediction that 40% of infrastructure and operations organizations scaling AI without strong governance will experience a business-critical disruption shouldn’t surprise anyone. Our own research found that 96.5% of organizations already have AI interacting with production databases, and we’ve already seen AI coding agents delete production databases. This isn’t a hypothetical problem anymore.

“Companies are giving AI more autonomy to make changes to the systems their businesses depend on, but the controls haven’t kept up. The issue isn’t just whether an agent has permission to access a database. It’s whether the changes it makes are safe. That means enforcing policies, understanding the potential impact before a change reaches production, and requiring human approval when the risk warrants it. AI may dramatically accelerate software delivery, but that speed means very little if it comes at the expense of production stability.”

In March 2026. Liquibase research found that 96.5% of organizations surveyed said AI already interacts with their production databases.

https://www.liquibase.com/2026-state-of-database-change-governance

Rakuten Securities breach exposes 15.7 million accounts

Posted in Commentary with tags on November 19, 2026 by itnerd

Japanese brokerage Rakuten Securities (Translation) has disclosed unauthorized access to an internal test environment that may have exposed customer information associated with up to 15.72 million accounts.

The company detected suspicious activity on November 16 and traced the potential data exposure to November 11.

The incident potentially affected 11.25 million accounts containing login IDs and customer codes, along with another 4.47 million accounts containing more sensitive information, including names, addresses, bank account details, annual income and financial assets.

Of particular concern, Rakuten acknowledged that it had temporarily copied real, unmasked customer data into its test environment to verify an internal system upgrade.

Ryan McCurdy, Vice President of Marketing, Liquibase

“Rakuten approved an exception to its normal security procedures and moved unmasked customer data into a test environment. The company had a legitimate reason to test a system upgrade, but that decision exposed sensitive information to unnecessary risk.

“Organizations must test changes before deploying them to production, but they shouldn’t have to put real customer information at risk to do it. Enterprises should mask sensitive data, restrict access to test environments, and enforce security policies that limit unnecessary exposure.

“This also raises a bigger question about how enterprises govern their database environments. An approval doesn’t automatically make an action safe. Organizations need to enforce consistent policies, control exceptions, and maintain audit evidence showing who authorized the exception, what data moved, and which safeguards were applied.”

ㅤ

John Strand, Owner, Black Hills Information Security

“Wow, what I find interesting about this incident is just how blunt the Japanese brokerage firm has been in describing what actually caused the problem and what they’re going to do moving forward. That’s refreshing. But once again, this really highlights the importance of basic security standards and practices.

“You do not use production data in test environments. Ever. Full stop. This isn’t some advanced security concept or something that requires sophisticated technology to address. It’s one of the most fundamental rules of protecting sensitive information. And yet, we continue to see organizations make these mistakes.

“The proper way to do this is to have separate development, testing, and staging environments, each with clearly defined security controls. Staging should closely mirror production in terms of configuration, infrastructure, access restrictions, and security protections, allowing teams to validate changes before deployment. But that doesn’t mean you should introduce live production data into staging. The standard practice is to use synthetic, masked, or properly anonymized data that accurately represents production conditions without exposing sensitive information. If real production data is absolutely necessary for a specific validation, that should be a tightly controlled exception, subject to formal approval, strict access controls, encryption, retention limits, and the same security protections required in production.

“There is a process for this. These development and testing practices have existed for well over a decade and a half. We’re not talking about inventing new security standards here. We’re talking about following established processes that companies should already have in place.

“And I think that’s what makes this incident so frustrating. These are basic security practices that need to be repeated again and again to development and testing teams. But I will give the brokerage firm credit for one thing. I’m very happy they were honest about what happened, acknowledged the problem, and explained what they’re going to do to fix it. We need more of that kind of transparency when security incidents occur.”

ㅤOrganizations need to adopt these things to protest themselves. Otherwise it will end badly.

New report urges stronger OT security as AI-powered attacks threaten critical infrastructure

Posted in Commentary with tags on November 19, 2026 by itnerd

The Operational Technology Cybersecurity Coalition (OTCC) has released a report warning that AI-enabled cyberattacks are accelerating threats to industrial controllers and other operational technology (OT) systems supporting U.S. critical infrastructure.

The November 17 report, The Missing Pillar of AI Security, outlines 18 recommendations for strengthening security and resilience across energy, water, manufacturing and other essential sectors.

The coalition cites earlier CISA warnings that attackers are using AI-generated scripts to target industrial controllers, exploiting familiar weaknesses such as internet-exposed devices and outdated software. The report calls for stronger secure-by-design requirements, protections for legacy and unpatchable OT systems, and improved cybersecurity funding for under-resourced infrastructure operators.

Citing a SANS Institute survey, the report notes that only 15% of respondents said their organizations could fully sustain manual operations beyond 48 hours. OTCC urges policymakers to prioritize resilience and strengthen cooperation between AI developers, government agencies and critical infrastructure operators.

John Strand, Owner, Black Hills Information Security

“One of the major concerns I have with these strongly worded reports is that absolutely everything in them is good advice. It’s great advice, in fact. Every organization should be trying to follow these recommendations. The problem is that we needed this type of guidance 15 or 20 years ago.

“I think one of the things that people really don’t understand, especially in the broader political and policy space, is just how complicated it is to deal with legacy technologies, particularly in operational technology and critical infrastructure environments.

“A lot of these systems simply cannot be replaced easily. In some cases, they can’t even be isolated from the rest of the network without creating serious operational problems. We’re talking about technologies that have been running for decades, supporting essential services, and were never designed with modern cybersecurity threats in mind.

“And that’s where I think a lot of these standards, recommendations, and guidelines fall short. They’re often developed in a vacuum, without a real understanding of what legacy critical infrastructure environments actually look like. It’s easy to tell an organization to segment its networks, replace unsupported systems, or implement modern security controls. It’s an entirely different thing to actually do that in an environment where taking a system offline could disrupt power, water, healthcare, or other essential services.

“The reality is that many of these organizations are holding their infrastructure together through an incredibly complicated collection of mitigating and compensating controls just to keep everything running safely and reliably. We absolutely need better security standards, but those standards also need to recognize the operational realities of the environments they’re supposed to protect. Otherwise, we’re just going to keep producing reports that tell organizations what they should have done 20 years ago without giving them a realistic path to improve security today.”

ㅤThis is another one of those required reading things. Because this is the way that the bad guys are going.