A lateral movement attack is the phase after initial access where an intruder moves through a network to reach valuable systems. It turns a single compromised host into a path toward an organization's critical data and assets.


What is a lateral movement attack?

Very fast. According to the CrowdStrike 2026 Global  Threat Report, attackers can begin moving laterally in as little as 27 seconds after gaining a foothold, and can reach the majority of an environment within an hour, far quicker than most detection processes respond.


How fast can lateral movement happen?

Lateral movement uses valid stolen credentials and built in system tools, so it looks like normal administrative activity. It generates no obvious anomalous signal, which lets it blend into routine network traffic and evade detection.


Why is lateral movement so hard to detect?

Stop lateral movement by limiting what compromised workloads can reach. Zero Trust segmentation, least privilege, multi factor authentication, and containment enforced at every workload close the network paths that lateral movement depends on.


How do you stop lateral movement?

Woman and man examining software code on a screen. Woman is pointing at the screen.

What a Lateral Movement Attack Really Costs You, and How to Shut It Down

Learn how a lateral movement attack spreads across networks, the business risks it creates, and how Zero Trust, segmentation, and Aviatrix help stop it.

Lateral Movement Attack: Costs, Risks & How to Stop It

Kubernetes security companies address container-specific threats: misconfigured security policies, exposed Kubernetes APIs, runtime threats inside running containers, lateral movement between workloads, and data exfiltration through ungoverned communication paths.


What do Kubernetes security companies actually protect against?

Kubernetes environments have unique characteristics: rapid pod churn that breaks IP-based security policies, east-west traffic between workloads that bypasses perimeter inspection, and ephemeral identities that require workload-identity-aware enforcement rather than static firewall rules.


How is Kubernetes security different from standard cloud security?

Kubernetes clusters run workloads that communicate continuously with other services and cloud APIs. Workload enforcement controls what those workloads can reach as they run. Without it, a single compromised container can move laterally across the entire Kubernetes environment before any scan or posture tool surfaces the issue.


Why does Kubernetes runtime protection matter for cloud security?

Ask whether the platform enforces communication policy or only identifies risk. Ask specifically how it governs east-west traffic between Kubernetes clusters, what coverage it provides on serverless and managed workloads without agents, and whether enforcement holds without requiring breach detection as a trigger.


What should I ask when evaluating Kubernetes security tools?

Top 8 Kubernetes Security Companies for 2026 Ranked

Top Kubernetes security companies in 2026, including Aviatrix, Wiz, Palo Alto, Sysdig, and CrowdStrike, ranked by what they actually enforce 


Export controls can create friction and delay, but history suggests they cannot permanently contain a capability once it has been publicly demonstrated. The encryption export controls of the 1990s, governed by the Export Administration Regulations and the Wassenaar Arrangement attempted to limit the spread of strong cryptography. Strong encryption spread globally anyway. AI is digital, borderless, and replicable. The containment challenge is not physical; it's epistemic. 

Do AI export controls work?

The Chokepoint Era describes the dominant security model of the past few decades, prevent threats from entering the environment through strong perimeters and access controls. The Containment Era is the emerging paradigm, driven by AI proliferation, that assumes capable technologies will reach environments regardless of restrictions. Security posture shifts from "keep it out" to "limit what it can do once it's in." 

What is the difference between the Chokepoint Era and the Containment Era in cybersecurity? 

Three steps:  

1. Audit what your AI agents can access, an agent with broad network and data access has a large blast radius if compromised. 
2. Design containment boundaries, treat AI workloads like you treat privileged accounts, with least-privilege access and lateral movement restrictions. 
3. Build detection for AI-specific threat vectors - prompt injection, model exfiltration, and agentic persistence are not covered by traditional endpoint security. 

What should organizations do to prepare for AI security risks? 

Why the Fable AI Ban Proves the Containment Era Has Arrived

The Fable AI ban didn't make organizations safer, it proved frontier AI can't be excluded. Learn why Containment Era security is the only viable response.

Fable AI Ban: You Can't Put the Genie Back in the Bottle

Project Glasswing is Anthropic's initiative to give cybersecurity defenders early access to Claude Mythos Preview — a frontier AI model capable of autonomously discovering and chaining vulnerabilities at a scale and speed no human team can match. For cloud security teams, Glasswing matters because it confirms that the window between vulnerability discovery and active exploitation has collapsed. Architectures built around detection can no longer respond fast enough. Containment — limiting Blast Radius before damage spreads — is now the primary security objective.

What is Project Glasswing and why does it matter for cloud security?

Glasswing shows that AI-speed exploitation targets the structural weaknesses already present in most cloud environments: flat east-west connectivity, overprivileged workloads, and identity sprawl across accounts and providers. These conditions give an attacker one reachable path and enough speed to move before a human analyst responds. The threat model shifts from "can we detect this?" to "how far can an attacker reach once they're inside?" Bounding Blast Radius through Communication Governance and runtime enforcement becomes the architecture priority, not faster alerting.

How does Project Glasswing change the threat model for cloud environments?

CNAPP and CSPM tools are Chokepoint Security instruments — they identify what could be exploited, but they do not enforce policy during active traffic. In an AI-speed threat environment, the constraint is no longer finding vulnerabilities. Attackers can move from initial access to lateral movement in minutes, faster than any human triage-and-escalation workflow. Runtime network-layer enforcement that isolates a compromised workload automatically is what limits damage. Posture scanning alone cannot contain an attack already in motion.

Why are Cloud Native Application Protection Platform (CNAPP) and Cloud Security Posture Management (CSPM) tools insufficient against AI-speed attacks?

Aviatrix Cloud Native Security Fabric addresses Glasswing's core gaps through three capabilities: runtime enforcement of Communication Governance policies at the network layer, stopping threats during active traffic rather than after the fact; lateral movement prevention by segmenting workloads within and across Virtual Private Clouds (VPCs) and Virtual Networks (VNets); and dynamic isolation via Aviatrix Distributed Cloud Firewall, which reduces containment time from hours to minutes. Policies are identity-based, not IP-based, so enforcement stays accurate as ephemeral infrastructure scales and changes.

What specific capabilities does Aviatrix provide to address the risks Project Glasswing highlights?

The immediate priority is shifting investment from detection toward survivability. That means implementing runtime workload segmentation so a compromised workload cannot freely reach others; establishing east-west visibility so teams know what normal communication looks like and can flag deviations; and replacing IP-based policy rules with workload identity-based enforcement that stays accurate as infrastructure changes. The operational question is no longer "how fast can we find threats?" It is "how quickly can we contain one?" Organizations that cannot answer that question in minutes need to reassess their architecture now.

How should CISOs prioritize cloud security investments in response to Project Glasswing?

Project Glasswing Didn-t Break Cybersecurity. It Exposed What Was Already Broken.

Project Glasswing revealed a hard truth: AI-powered attacks move faster than remediation. Here's what cloud security teams must change now. 


Project Glasswing Exposed Cloud Security's Biggest Weakness

Project Glasswing Didn't Break Cybersecurity. It Exposed What Was Already Broken.

The Anthropic Claude AI vulnerability allowed attackers to steal user chat data without using malware, phishing, or traditional hacking methods. The attack chained together three weaknesses: hidden prompt injection through URLs, abuse of the Files API to move data, and an open redirect vulnerability to exfiltrate information. The AI processed malicious instructions embedded in harmless-looking content and could not distinguish between legitimate user requests and attacker commands. Once manipulated, Claude accessed sensitive data and moved it through trusted application flows, making the attack nearly invisible to conventional security monitoring tools.

What is the Anthropic Claude AI vulnerability, and how did it work?

This AI security risk is significant because the attack succeeded from inside a trusted system. Claude acted with valid credentials and followed legitimate access paths, so standard defenses like endpoint protection and identity-based access controls did not flag anything unusual. Traditional security models assume users initiate actions, applications follow predictable logic, and threats come from outside the network. Agentic AI breaks all three assumptions. The AI became the attack vector itself, executing authorized actions in the wrong context. This signals a broader shift in how security teams need to think about autonomous software operating inside enterprise environments.

Why does the Anthropic Claude AI vulnerability represent a new kind of AI security risk?

Security teams can reduce AI security risk by enforcing controls at the network layer rather than relying solely on application-level defenses. Key steps include restricting outbound traffic to approved destinations, segmenting AI workloads from sensitive data stores, and applying micro-segmentation to separate inference, storage, and integration paths. Visibility into AI-driven network behavior is also critical. When all traffic looks like valid API calls, flow-level monitoring and behavioral baselines help detect anomalies. Assuming AI can be manipulated and architecting systems to contain the damage is a more realistic security posture than trying to prevent every possible prompt injection attack.

How can security teams reduce AI security risk from vulnerabilities like the one found in Anthropic Claude?

The Anthropic Claude AI vulnerability reveals that AI agents must be treated as a distinct security domain. Organizations have built strong defenses around users, endpoints, and applications, but autonomous software entities that can reason, act, and move data require a different approach. These agents are frequently over-permissioned, with broad access to internal storage, APIs, and external services simultaneously. Security teams need to apply least-privilege principles to AI workflows, enforce strict egress controls, and deploy cloud network security tools that can observe and limit what a compromised AI agent can actually do once it has been manipulated.

What does the Anthropic Claude AI vulnerability reveal about securing AI agents going forward?

Lessons from the Claude Data Exfiltration Vulnerability

Ian Smith

Ian Smith is a Senior Principal Solution Architect at Aviatrix.

Senior Principal Solution Architect

Jason Haworth

Jason is an experienced leader and technologist helping companies build great teams and culture.

Principal Solutions Architect, CNSF

Recently-disclosed vulnerabilities in Anthropic’s Claude AI shows how AI can become an insider threat. Learn the risks and what security teams can do.

What the Claude Data Exfiltration Vulnerability Teaches Us 

The Architectural Divide is the gap between modern cloud applications and the security tools meant to protect them. Applications today are spread across many clouds and change quickly. The security controls around them are broken up by cloud provider, often disconnected from the actual flow of traffic. This leads to three key problems: fragmentation, where policy must be rewritten for each cloud; unclear ownership, where no one team controls the full data path; and weak runtime enforcement, where tools can show what happened but cannot stop threats as they move between workloads in real time.

What is the Architectural Divide?

Aviatrix closes the Architectural Divide by tackling its three core gaps. It builds a single control plane across clouds so policy is written once and enforced everywhere, using metadata like tags and labels instead of shifting IP addresses. It gives teams true ownership of the data path by putting policy-aware enforcement straight into the cloud data plane, with policy-as-code so new workloads get secured without waiting on a human. And it embeds enforcement into the network fabric itself, stopping lateral movement at the network layer before it becomes a breach rather than flagging it after the fact.

How does Aviatrix help close the Architectural Divide?

Cloud Native Security Fabric (CNSF) is how Aviatrix brings scattered security tools into one working whole. Instead of stacking more point products on top of broken-up cloud setups, CNSF gives you a steady enforcement backbone: policy defined once, enforcement spread everywhere, visibility linked across clouds, segmentation held firm across workload types, and ownership grounded in the architecture itself. CNSF does not tear out your current security tools. It ties them together and keeps enforcement steady from Layer 3 through Layer 7, helping teams shrink blast radius by design and speed up secure development across clouds.

What is Cloud Native Security Fabric?

How Aviatrix Helps Bridge the Architectural Divide

The Architectural Divide could unravel cloud network security. Learn how Aviatrix CNSF bridges the divide through unified visibility and policy enforcement.

Bridging the Architectural Divide with Aviatrix

The Architectural Divide

Learn about the architectural divide in modern cloud security between how applications are built and security is enforced. Only CNSF can close this gap.

The Architectural Divide in Cloud Security

The Architectural Divide: Why Modern Cloud Security Falls Behind

Cloud Network Security, Simplified The High-Rise Apartment Analogy

Cloud Network Security, Simplified The High-Rise Apartment Analogy card image

Cloud network security needs a new model. Learn how through the analogy of an apartment building’s security upgrades. 

What is Cloud Network Security? An Analogy

Cloud Network Security, Simplified: The High-Rise Apartment Analogy

What Caused the AWS Outage And How to Prevent Similar Incidents card image

What Caused the AWS Outage And How to Prevent Similar Incidents

The AWS outage on October 20-21 was caused by DNS resolution failures to DynamoDB in us-east-1. Learn what happened and how to prevent future cyber incidents.

The AWS Outage Explained: Root Cause, Lessons & Prevention

What Caused the AWS Outage? And How to Prevent Similar Incidents

Legacy Tools Fall Short Operationalizing Zero Trust for Your Security Stack

Legacy Tools Fall Short Operationalizing Zero Trust for Your Security Stack v4

CNAPP, CSPMs, EDRs, SASE, SSE, NGWs, and other security tools fall short of runtime security. Learn how a cloud native security fabric (CNSF) fills the gaps.

Jason Haworth

Author Biography

Published Articles by Jason Haworth

What a Lateral Movement Attack Really Costs You, and How to Shut It Down

Top 8 Kubernetes Security Companies for 2026 Ranked

Why the Fable AI Ban Proves the Containment Era Has Arrived

Project Glasswing Didn't Break Cybersecurity. It Exposed What Was Already Broken.

Lessons from the Claude Data Exfiltration Vulnerability

How Aviatrix Helps Bridge the Architectural Divide

The Architectural Divide: Why Modern Cloud Security Falls Behind

Cloud Network Security, Simplified: The High-Rise Apartment Analogy

What Caused the AWS Outage? And How to Prevent Similar Incidents

Legacy Tools Fall Short: Operationalizing Zero Trust for Your Security Stack

Products

By Role

Customer Stories

Security & Legal

Solutions

Resources

Industry

Company

Tools