News

Is It AI? After Nationwide Outages, Broadcom and CISA Sound Alarms on Critical Security Flaws

After a string of nationwide technology outages affecting airlines, government systems, banking and cloud services, urgent new warnings from Broadcom and the Cybersecurity and Infrastructure Security Agency (CISA) are raising an increasingly relevant question: Is artificial intelligence contributing to the growing turbulence in enterprise technology infrastructure?

The possible connection isn't entirely speculative. The Cloud Security Alliance (CSA) has examined whether AI-assisted vulnerability discovery and accelerated security patching could be contributing to widespread disruptions. Although its Nov. 18 investigation found no evidence connecting AI-related patching to the recent outages, it identified growing pressure on enterprise change-management processes as a legitimate concern.

Against that backdrop, Broadcom has disclosed a critical VMware vCenter vulnerability exploited in attacks, while CISA has added four actively exploited vulnerabilities to its catalog. Neither announcement has been linked to the outages, but both underscore the security pressures facing enterprise IT teams.

Broadcom Warns of Exploited VMware Zero-Day
Broadcom on Nov. 19 issued security advisory VMSA-2026-0009, addressing five vulnerabilities affecting VMware vCenter, VMware Tools, SDDC Manager and related virtualization platforms.

The most serious, CVE-2026-123890, is a VMware Directory Service authentication bypass rated 9.8 on the CVSS severity scale. An attacker with network access to an affected vCenter server could exploit the flaw to bypass authentication and obtain unauthorized access.

The vulnerability is particularly dangerous because vCenter provides centralized control over VMware virtualization environments, including the ability to configure and administer ESXi hosts and virtual machines.

Google Threat Intelligence Group and Mandiant reported that a suspected China-linked espionage group tracked as ARMORY CASTLE exploited the vulnerability as a zero-day in September.

According to the Google/Mandiant investigation, the attackers used a specialized Golang tool called BONESAW to exploit vCenter authentication mechanisms, obtain sensitive database information and establish administrative access.

The researchers also documented additional malware designed for credential theft, persistence and lateral movement through compromised infrastructure.

Broadcom has released fixes for supported product versions, including vCenter 8.0 U3m and updates for VMware Cloud Foundation 9.x and 5.x. Customers running vCenter 7 under extended support must contact Broadcom for remediation guidance.

Broadcom lists no workaround for the critical authentication bypass, making the applicable software updates the primary remediation.

The advisory also addresses SQL injection and command injection in SDDC Manager, along with additional security weaknesses affecting VMware infrastructure components.

CISA Adds Four Exploited Vulnerabilities
A day earlier, CISA updated its Known Exploited Vulnerabilities catalog with four vulnerabilities affecting Apache Axis, BerriAI LiteLLM, Vanderbilt University's REDCap and Synology DiskStation Manager.

Inclusion in the catalog means CISA has identified evidence of exploitation in the wild. It does not establish that the four vulnerabilities are connected to a single attack campaign or to the recent service outages.

The additions include:

  • CVE-2026-90436: An Apache Axis vulnerability involving unsafe class loading that could permit unauthenticated remote code execution under affected configurations. Apache Axis 1.x is end of life.
  • CVE-2026-40217: A LiteLLM vulnerability allowing attackers who can access the affected custom-code guardrail testing functionality to escape its Python sandbox and execute arbitrary code.
  • CVE-2026-90817: A REDCap code injection vulnerability that could allow unauthorized server-side code execution through specially constructed requests, provided an attacker knows a valid public survey hash.
  • CVE-2024-10441: A Synology vulnerability affecting the system plugin daemon that could permit remote code execution.

The LiteLLM finding is particularly relevant to organizations deploying AI infrastructure. LiteLLM is used as a gateway for accessing and managing large language models, and its affected custom-code testing endpoint could permit execution outside the intended sandbox.

The project's security advisory recommends upgrading to version 1.83.11 or later. Administrators unable to upgrade immediately can block access to the affected testing endpoint.

CISA's listed remediation deadlines for covered federal agencies are Nov. 21 for Apache Axis, REDCap and Synology, and Dec. 2 for LiteLLM. Private organizations should evaluate affected deployments according to their own exposure and risk.

Is AI Contributing to the Disruptions?
The question raised by the timing of the outages and security alerts is whether AI is introducing new instability into increasingly complex enterprise infrastructure.

The CSA researchers cautioned against allowing speculation about AI to get ahead of incident investigations. "adversarial AI has become the default hypothesis for any unexplained failure," they wrote, warning that premature assumptions about either an attack or an innocent technical failure can undermine incident response.

In its Nov. 18 analysis, the CSA AI Safety Initiative examined disruptions between Nov. 12 and Nov. 17 affecting Microsoft 365, Workday, the FAA, AWS, financial institutions, airlines, Apple and the National Security Agency.

The organization investigated a theory that AI-powered security tools are discovering vulnerabilities faster than organizations can safely respond. Under that theory, an increasing volume of emergency patches could introduce defects or operational failures into interconnected systems.

It is a plausible concern about the industry's growing reliance on AI-assisted vulnerability discovery and increasingly automated software maintenance. But the investigation found no evidence that such patching caused the recent disruptions.

Instead, the documented incidents involved a variety of more conventional causes, including software defects, network configuration changes, routine maintenance and problems involving recovery or failover operations.

For example, Microsoft's outage involved a previously introduced network-related defect that surfaced under later operating conditions. The FAA attributed its nationwide ground stop to unstable legacy equipment overwhelmed by connection attempts associated with networking changes.

These findings do not rule out AI-related operational risks in general. They do, however, undermine the suggestion that the recent outages share a demonstrated AI cause.

More Vulnerabilities, Faster Responses, Greater Risks?
The CSA analysis points to a broader concern: As AI tools accelerate vulnerability discovery, enterprise security teams may face increasing pressure to evaluate, test and deploy fixes across complex infrastructure environments.

That pressure creates a difficult operational balance. Delaying a critical patch can leave infrastructure exposed to attackers, particularly when exploitation is already documented. Deploying changes without sufficient testing can introduce availability problems of its own.

Broadcom's vCenter advisory illustrates the first risk. A critical vulnerability in a central virtualization management system was exploited before its public disclosure, requiring administrators to respond to an immediate security threat.

The recent outages illustrate a different challenge: Changes or failures within shared infrastructure can propagate into services used by large numbers of organizations.

There is no verified evidence that AI caused either the outages or the vulnerabilities highlighted by Broadcom and CISA. But the combination underscores an emerging question for enterprise IT: Can traditional security and change-management processes keep pace as AI accelerates the discovery of weaknesses and potentially the development of exploits?

For now, the answer to whether AI was behind the recent nationwide failures remains unsupported by the available evidence. The larger challenge of maintaining secure, reliable infrastructure while responding to increasingly rapid technological change remains very real.

About the Author

David Ramel is an editor and writer at Converge 360.

Featured