June 15, 2026

Hola VPN Abuse: From Proxy Traffic to Malware and Cryptomining

Darktrace’s analysis of Hola VPN-related activity across multiple environments reveals how peer-to-peer proxy functionality can enable malware delivery, command-and-control traffic, and cryptomining. Darktrace data highlights consistent patterns, suspicious downloads from rare endpoints, and post-compromise behavior, demonstrating how seemingly benign software can obscure threats and facilitate further malicious activity.

Written by

Min Kim

Cyber Security Analyst

Inside the SOC

Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.

Written by

Min Kim

Cyber Security Analyst

Jun 2026

Introduction

In enterprise environments, non-compliant software traffic can introduce unexpected exposure by creating unmanaged paths for outbound connectivity. Hola VPN is a notable example because of its peer-to-peer design, which can effectively turn user devices into routing or exit nodes for other parties’ traffic, shifting the risk profile from that of a traditional virtual private network (VPN) to something closer to a distributed proxy.

As a result, the appearance of Hola-related activity, whether from prior installation or unintended background connections, should be treated with caution. Such activity may provide a foothold for malicious behavior, including lateral movement or command-and-control communication.

This blog explores how Hola-associated activity appeared as part of broader patterns of suspicious behavior observed across the Darktrace customer base.

The campaign

In February and March 2026, Darktrace observed similar anomalous activity across multiple customer environments, with affected devices showing consistent behavioral patterns. These included connections to multiple *.hola[.]org endpoints using Hola-related user agents, suggesting interaction with Hola infrastructure rather than isolated or incidental traffic.

Following these connections, affected customer environments showed downloads of suspicious executable files from rare external endpoints 188.241.219[.]55 and 184.241.218[.]111. Both endpoints have been flagged as potentially malicious by open-source intelligence (OSINT) [1][2].

These downloads were conducted using consistent user agents across impacted customers, specifically ‘Hola svc_js_win32/1.249.408’ and ‘Hola svc_js_win32/1.251.389’, suggesting a possible association with Hola-related activity.

Notably, this pattern aligns with recent reporting that, in some cases, Hola distributed an undeclared executable component, me[.]exe, which was later assessed to be a likely Monero-mining binary introduced via a compromised delivery pipeline [3].

Case Study 1

Darktrace first observed a new device on January 19, 2026, within a customer environment based in the Europe, Middle East, and Africa (EMEA) region. On the same day it appeared on the network, the device communicated with multiple pieces of Hola VPN-linked infrastructure before downloading a binary from a hola[.]org subdomain.

Figure 1: Cyber AI Analyst investigation highlighting Hola VPN service activity potentially associated with subsequent HTTP command-and-control (C2) connections.

Subsequent Darktrace telemetry revealed a recurring pattern of activity from the day the device was first observed through to March 4, 2026. During this period, the device repeatedly issued HTTP GET requests to the URI /bwfile?size=1048576, each returning a 200 OK response, indicating successful file retrieval.

This behavior was accompanied by a POST request to /bwfile, followed by an additional GET request for a significantly larger file at /bwfile?size=26214400, suggesting a deliberate and structured file transfer pattern.

Notably, the binary download activity was not tied to a single static host. Instead, it was observed across multiple URLs that changed over time while remaining within the same hola[.]org domain. This pattern suggests the use of rotating or distributed delivery infrastructure rather than a fixed endpoint.

Figure 2: Variation in URLs over time within the same hola[.]org domain, indicating the use of dynamically changing endpoints.

Across these events, the activity was consistently associated with the user agent Hola svc_js_win32/1.249.408, further linking the traffic to Hola-related service components. Amid these persistent and unusual connections, on February 22, Darktrace observed the device connecting to 188.241.219[.]55/proxy-peer-windows-amd64[.]exe, resulting in the download of an executable file.

Figure 3: File transfer event showing the download of an executable from the rare external endpoint 188.241.219[.]55.

Based on its file hash, the downloaded file was assessed as a likely Trojan downloader [4], with import hash (imphash) values showing similarities to samples linked to Vidar, Rhadamanthys, and Stealc according to OSINT [5]. Overall, this sequence of activity suggests that Hola-related connectivity may have been leveraged as part of a broader malware delivery chain.

Darktrace’s Autonomous Response

Due to the highly unusual activity observed, Darktrace Autonomous Response was triggered by the device’s behavior. However, as the customer deployment was configured in “Human Confirmation” mode, manual approval was required before any action could be taken.

Had the deployment been set to “Fully Autonomous” mode, Darktrace would have automatically:

Blocked connections to the associated ports and external endpoints
Prevented all outgoing network connections from the device
Enforced the device’s established ‘pattern of life’, allowing normal activity to continue while restricting any anomalous behavior

Figure 4: Example of a Darktrace Autonomous Response model highlighting the action that would have been taken, demonstrating how the system identifies anomalous behavior and applies targeted containment measures to restrict suspicious network activity.

Case Study 2

While the first case focused on anomalous activity from a newly observed device, Darktrace also identified cases in which devices had already been communicating with Hola-related endpoints prior to the suspected campaign. This may suggest pre-existing Hola usage within the environment, potentially increasing exposure and creating an avenue for subsequent suspicious activity.

One case involved three devices within a customer network based in the Americas (AMS). In this instance, a different payload was identified: me[.]exe, a potentially malicious cryptocurrency miner also referred to as HolaMonitorService[.]exe [6][7]. The downloads were observed from infrastructure similar to that seen in Case 1, including an IP address within the same 188.241.0.0/16 subnet.

Connections to *.hola[.]org, alongside the use of potential Hola-related user agents consistent with those in Case 1, were also identified, further suggesting a link between the observed activity and Hola-associated infrastructure.

Darktrace observed activity indicative of unusual VPN usage on the first affected device on February 2, followed by telemetry suggesting potential Tor usage. This was later followed by the download of me[.]exe on March 10 from 188.241.218[.]111. Notably, this device was the earliest among the three within the deployment to exhibit the presence of the suspicious executable.

Figure 5: Cyber AI Analyst detection highlighting the download of a suspicious executable from a similar external endpoint in a separate deployment.

On March 5, 2026, the second affected device exhibited a slightly different progression, initiating connections to http-test1[.]hola[.]org using the user agent ‘hola_get’. This activity was followed by the download of me[.]exe from the same endpoint on March 13, consistent with the broader pattern of Hola-related downloads observed across the environment.

Figure 6: Example of Hola VPN-related connectivity observed on the network prior to the suspected campaign, indicating pre-existing usage that may have contributed to subsequent activity.

The final affected device within this customer’s network demonstrated a more limited but related pattern, also downloading me[.]exe on March 17 using the same ‘hola_get’ user agent.

While the earlier Hola VPN usage observed across the deployment may not have been directly related to the suspected malware campaign, it may nonetheless have contributed to reduced visibility. The presence of pre-existing Hola-related traffic could have obscured malicious activity, making it more difficult to distinguish legitimate usage from attacker-driven behavior and, in turn, hindering the timely identification of the emerging compromise.

Darktrace’s Autonomous Response

For this deployment, the customer had their Autonomous Response capability configured in “Fully Autonomous” mode, allowing Darktrace to take action without human intervention. As a result, the system was able to autonomously disrupt the activity as soon as relevant events were identified through model detections.

Figure 7: Darktrace Autonomous Response actions taken against suspicious activity linked to Hola VPN.

Suspected cryptomining activity

As previously noted, some of the observed executable payloads appear to be linked to cryptomining malware. Across a subset of affected customer environments, this assessment was further supported by subsequent device activity consistent with Monero mining. Affected devices established follow-on connections to multiple external endpoints aligned with known mining infrastructure, indicating post-download execution.

Considering the broader sequence of activity, this pattern may point to a wider form of abuse in which legitimate VPN-related traffic is used to mask or facilitate malicious behavior following compromise.

On several devices, the download of executable files, including a newly observed peer[.]exe, was followed by alerts indicative of cryptocurrency mining activity. Mining-related credentials such as ‘x’ were observed using the Minergate protocol to communicate with endpoints within the 89.125.255.0/24 subnet and 188.241.218[.]111, the same endpoint involved in earlier download activity. Additional credentials appeared to reflect device-specific CPU identifiers, for example ‘12th Gen Intel(R) Core (TM) i5-1235U’.

Observed mining methods included login, submit, and job, consistent with active participation in a pool-based mining workflow rather than passive or incidental contact. The login method indicates that the host authenticated to the mining service as a worker, job reflects the assignment of computational tasks, and submit shows completed work being returned to the pool [8]. This sequence suggests that affected devices were actively contributing processing resources as part of an unauthorized distributed mining operation.

The presence of unauthorized cryptominers can lead to degraded system performance and reduced device stability. Beyond the immediate resource impact, such activity often serves as an indicator of a broader compromise rather than an isolated issue. This may increase the risk of further malware deployment, persistence mechanisms, and lateral movement, particularly in environments where the initial intrusion has not been fully contained.

Conclusion

Across affected environments, detections such as unusual VPN usage, connections to Hola infrastructure, anomalous HTTP activity, suspicious file downloads, and subsequent cryptomining behavior were linked into a single, evolving incident narrative. This aggregation provided a clearer view of attack progression, enabling security teams to understand not just isolated alerts, but the full sequence of compromise from initial contact through to post-exploitation.

Ultimately, these activities show that the risk posed by non-compliant software such as Hola VPN can extend far beyond simple policy violations. What began as traffic to Hola-related infrastructure was, in multiple cases, followed by behavior suggesting deliberate misuse, including suspicious executable downloads using Hola-related user agents and, in some instances, evidence of active cryptomining. These were not isolated anomalies, but elements of a broader pattern in which seemingly benign proxy or VPN-related communications may have created a pathway for malicious delivery and unauthorized resource exploitation.

The significance of this activity lies not only in the downloads or mining, but in what it reveals about an attacker’s ability to blend malicious operations into traffic associated with software that may already have a foothold in the environment. When unapproved software operates within an enterprise, it can reduce visibility, blur the distinction between legitimate and malicious traffic, and create opportunities to extend compromise in ways that are persistent and difficult to detect. Darktrace’s anomaly-based approach enables these behavioral distinctions to be identified, regardless of whether the device is new or long established within the network.

Credit to Min Kim (Associate Principal Analyst), Priya Thapa (Senior Cyber Analyst)
Edited by Ryan Traill (Content Manager)

Appendices

References

[1] https://www.virustotal.com/gui/ip-address/188.241.219.55

[2] https://www.virustotal.com/gui/ip-address/188.241.218.111

[3] https://www.sophos.com/en-us/blog/you-do-surprise-me-exe-an-unexpected-executable-in-hola-browser

[4] https://www.virustotal.com/gui/file/d275abca286cd75af971d0459fdf1df37c7b19c514abafae5d0b04bf42ccfb45/detection

[5] https://bazaar.abuse.ch/sample/d275abca286cd75af971d0459fdf1df37c7b19c514abafae5d0b04bf42ccfb45/

[6] https://any.run/report/4cdeb5df217764a8b6a20d518b76ccb30cbe623365a13d9dcd40900950f1ed99/de3a756a-3101-4369-8922-52c586c939fb

[7] https://www.virustotal.com/gui/file/e3541caf708c075f0bb22fc68b03acd8457fea7cf0732ea935b1eb016d1c7721/community

[8] https://bitcoinwiki.org/wiki/stratum

Darktrace Model Detections

· Anomalous File / EXE from Rare External Location

· Anomalous File / Multiple EXE from Rare External Locations

· Compromise / Crypto Currency Mining Activity

· Compromise / High Priority Crypto Currency Mining (EM)

· Device / New User Agent

· Anomalous Connection / New User Agent to IP Without Hostname

· Antigena / Network / Significant Anomaly / Antigena Controlled and Model Alert

· Antigena / Network / Significant Anomaly / Antigena Alerts Over Time Block

· Antigena / Network / External Threat / Antigena Tor Block

· Antigena / Network / External Threat / Antigena File then New Outbound Block

· Antigena / Network / External Threat / Antigena Suspicious Activity Block

· Antigena / Network / External Threat / Antigena Suspicious File Pattern of Life Block

· Antigena / Network / External threat / Antigena Suspicious File Block

Indicators of Compromise (IoCs)

IoC –Type -Description + Confidence

188.241.219[.]55 - IP Address - Malware distribution source

188.241.218[.]111 - IP Address -Malware distribution source

hxxp://188.241.218[.]111:8080/me[.]exe - URI - Malicious payload

hxxp://188.241.219[.]55:9000/proxy-peer-windows-amd64[.]exe - URI - Malicious payload

hxxp://188.241.219[.]55:9000/peer[.]exe - URI - Malicious payload

C8088f3c8bc3542eb1ad78a7cc5306d866c8ac81 - SHA1 - Malicious payload, me[.]exe

b595a6de0f6a18975b29e6f8ebe604956a173478 - SHA1 - Malicious payload, me[.]exe

e9139a2e0839e8b9e5c9787ea936347ae56e5460 - SHA1 - Possible malicious payload

c2e80073e4cafe757d5643bd8fd45f28ad89bff9 - SHA1 - Possible malicious payload

695355eceedcdd337d8fcbd35e6a531cda75b847 - SHA1 - Possible malicious payload

f0b0d8068a1b9ab5d68a8a46842d72b870b292e7 - SHA1 - Possible malicious payload

a21c8b8cabc7670ea45bc175e185a0f9bfcf4733 - SHA1 - Malicious payload, me[.]exe

0353ca44b9f397d8f492db0b2f7a1d00a9e4406a - SHA1 - Possible malicious payload

56824c8a110e35ab303dc27a6c758cd50c36174c - SHA1 - Malicious payload, peer[.]exe

c141fa0fa505fe7f9ad5dd21d9d4d6d411739682 - SHA1 - Malicious payload, peer[.]exe

0417ec988b16f1267065185a6eea98f0bd2e17cd - SHA1 - Possible malicious payload

c54f7eaaeb3e0b528cd2584bdcb3a4b13cc0f8a2 - SHA1 - Malicious payload, peer[.]exe

11c78f15fafd53f8cc5a52b828d7cbf2a99e0b09 - SHA1 - Malicious payload, peer[.]exe

0258bf7dbb0123247db29e8799991140bbdbd9bb - SHA1 - Malicious payload, proxy-peer-windows-amd64[.]exe

b46043a06dd9bbd63e4214d5fbc7fd56e1ff0618 - SHA1 - Possible malicious payload

753afdecd9f5402d004e8e5f768170ae9a468ca5 - SHA1 - Possible malicious payload

8f533c7cb1524b00f7b0311c2ea8603298d6b2ca - SHA1 - Possible malicious payload

3a3bc6a5b4db1a4e961abcb002d26fe9d5e5c349 - SHA1 - Possible malicious payload

897f70eb41d302b045fcb05ed0693675e778ce57 - SHA1 - Possible malicious payload

6ddd5644809606e3dc1e2cc06059c3f5e6176f85 - SHA1 - Malicious payload, proxy-peer-windows-amd64[.]exe

68a94f7cdcaf8853ea99251c1ecc67ae9b32eba8 - SHA1 - Malicious payload, proxy-peer-windows-amd64[.]exe

MITRE ATT&CK Mapping

T1659 -Initial Access, Command and Control -Content Injection

T1588.001 -Resource Development -Malware

T1189 -Initial Access -Drive-by Compromise

T1105 -Command and Control -Ingress Tool Transfer

T1657 -Impact -Financial Theft

T1497.001 -Impact -Compute Hijacking

T1496 -Impact -Resource Hijacking

T1210 -Lateral Movement -Exploitation of Remote Services

T1036.012 -Stealth -Browser Fingerprint

T1071.001 -Command and Control -Web Protocols

Written by

Min Kim

Cyber Security Analyst

Inside the SOC

Written by

Min Kim

Cyber Security Analyst

Watch the NIS2 Webinar

Blog

July 7, 2026

Securing AI: Analysis of the Complete Security Stack with Governance and Controls

Why traditional cybersecurity approaches are not enough for AI

AI adoption outpaces most security programs’ ability to adapt. That gap is now one of the most consequential sources of cyber risk facing enterprises. As organizations embed generative and agentic AI into development workflows, business operations, and security tooling itself, the question is no longer whether AI will introduce risk. The question is whether organizations understand where that risk actually lives and how to manage it operationally.

Two recent pieces of guidance underscore this shift:

The upcoming Cybersecurity Framework Profile for AI from NIST

The Five Eyes government guidance on the careful adoption of agentic AI services

Taken together, they point to a critical conclusion. AI security cannot be reduced to model hardening or prompt filtering. It requires a defense in depth strategy that treats AI as both a new attack surface and a force multiplier for defense, while accounting for how AI fundamentally changes scale, speed, and autonomy.

Recent threat research suggests that today's cyber risk is driven less by initial compromise and more by an adversary's ability to blend into normal operations over time. AI systems create the same exposure in a new form: more autonomy, more scale, and more opportunities for risky behavior to blend into normal operations.

How NIST defines the three core pillars of AI security

The NIST profile organizes AI risk across three inseparable focus areas that span all cybersecurity functions, Secure, Defend and Thwart. These areas are not sequential. They exist simultaneously and must be addressed together.

Secure

This treats AI as an attack surface. It includes models, prompts, agents, pipelines, training and inference data, retrieval augmented generation corpora, and the AI supply chain itself. AI systems are opaque, probabilistic, and non-deterministic by design. Some vulnerabilities are inherent in how models are trained or how data is sourced. Traditional patching does not fully mitigate these risks. This is also where many enterprises are weakest today and, critically, where many security programs stop.

Defend

This is AI as a defensive force multiplier. AI can improve detection speed, scale, correlation, and response, but only if the right models are used and operationalized correctly. Machine-speed behavior-based detection, response and containment becomes critical in defending non-deterministic systems. Accuracy, explainability, governance, testing, validation, and integration into SOC workflows matter as much as capability. Without those controls, hallucination risk, over automation, and misplaced trust become security risks themselves.

Thwart

This treats AI as an adversarial accelerant. Threat actors are already using AI to generate targeted social engineering attacks, deepfakes, malware, and autonomous attack agents. Asymmetric warfare is highlighting faster vulnerability discovery and exploitation with a lag on patch development, testing and deployment.

How this looks in practice

Darktrace researchers observed scaled, automated exploitation of the React2Shell vulnerability within days of disclosure. A vulnerable cloud asset was exploited in under 120 seconds of being deployed. Darktrace research team observed an AI/LLM-generated malware sample used in exploitation activity tied to React2Shell. The significance isn't novelty. It is that AI lowers the barrier to producing usable offensive tooling and compresses the time between experimentation and deployment.

Tactics are getting more and more creative in order to string together steps of an attack kill chain. This creates a dependency on behavior-based detection, autonomous investigation, autonomous containment, training, resilience investment, and recovery planning across the entire enterprise.

Why agentic AI fundamentally changes enterprise cyber risk

The Five Eyes guidance on agentic AI highlights material changes to the cyber risk profile of an organization. Unlike generative AI systems that produce content for human consumption, agentic AI systems reason, plan, and act autonomously across tools, data, and environments. That autonomy, combined with access to real systems, amplifies the impact of traditional cyber failures and introduces new system level risks that are difficult to predict, observe, and contain.

Risk in agentic systems does not live in the model alone. It emerges from interactions between models, prompts, memory, tools, APIs, identities, privileges, inter-agent trust relationships, and human assumptions baked into design. Vulnerabilities are often introduced through data, connectors, natural language interfaces, protocols, and drift by design.

In supply-chain incidents, attackers did not need sophisticated exploits to scale impact. They abused trusted systems built for automation and implicit access. Agentic AI inherits that model. Once a system can act across tools, data, and workflows, compromise propagates through trust relationships that were never designed for machine autonomy.

The major agentic AI risk classes include the following:

The identity control for non-human identities or autonomous agents makes it difficult to mitigate over-permissioning, limiting access, scope, and duration, as well as access hygiene

Agents are frequently over permissioned

Compromised tools inherit agent authority

Static secrets enable impersonation

Implicit trust between agents enables lateral movement

Design and configuration risks compound this, including privileges evaluated once at startup, poor segmentation, unvetted third party tools, reused authorization decisions outside their original context, and guardrail limitations.

Behavioral risk

Agents can optimize for goals in unsafe ways, misinterpret ambiguous intent, chain actions into unintended sequences, change behavior during evaluation, and exhibit deceptive or sycophantic responses.

Structural risk

Structural risk follows from agentic systems that are tightly coupled, multicomponent ecosystems. Failures can propagate across agents. Hallucinations cascade downstream. Resource exhaustion becomes systemic. Tool misuse enables indirect prompt injection and command execution. Rogue agents can poison peer agents through trust relationships.

Accountability

Accountability becomes unclear as autonomy increases. Autonomous agents assume human identity permissions, and humans should have clear ownership of these agents, but they don’t, and this model is flawed. Decision paths are opaque and non-deterministic. Logs are fragmented and difficult to interpret. Reproducing an incident will be impossible without explicit design for observability and forensics. An agent compromise is functionally an insider threat, often with better access and fewer behavioral constraints than a human.

What does defense in depth look like for AI?

Agentic AI runs on software, networks, identities, and data. It must be governed using the same foundational principles that have proven resilient under uncertainty, including secure by design, defense in depth, zero trust, least privilege, continuous monitoring, behavior-based advanced threat detection and containment, and incident response and recovery.

Core components to a Defense in depth Strategy for Securing the use of AI:

Strong, precise identity control plane to include an identity per agent (cryptographic, non‑shared)
- Privilege monitoring and just‑in‑time access

Data Governance

Secure‑by‑default configurations
- Security Posture Management
- Zero Trust principles

Strong guardrails, deny‑by‑default policies, and isolation

Explicit instruction hierarchies and controlled context

Behavioral-based detection across entire enterprise to include inputs, tools, and outputs as well as AI used on the endpoint, across the network, cloud, SaaS, email, and OT
- Runtime anomaly detection and goal‑drift detection
- Autonomous containment to mitigate risk and minimize damage

Hard boundaries on autonomy and delegation

Testing, Evaluation, Validation and Verification
- Determine when autonomous action and when human in the loop
- Adversarial training and agent‑specific testing
- Simulation, red teaming, and chaos testing

Kill‑switches, rollback, and containment mechanisms
- Forensics data captures, interpretability, autonomous containment, and remediation/recovery plans

Until standards, tooling, and assurance methods mature, organizations should assume agentic AI systems will behave unexpectedly and design deployments around resilience, behavior-based detection, reversibility, and containment, not efficiency.

How security leaders should prepare for enterprise AI adoption

AI security is not model security alone. Data, pipelines, identities, and agents are first class assets. Many AI attacks succeed through standard cyber failures amplified by AI. Identity, data, and supply chain risk dominate. Behavior-based detection and response are critical, not optional. Logging, provenance, versioning, and forensics data capture of detections are mandatory because you cannot investigate or recover from AI incidents without them.

Risk will often be visible in behavior before it is clearly defined in policy or guidance. The same pattern has been seen in pre-CVE disclosure detection, where abnormal activity appears before the industry has named or described the vulnerability. AI systems introduce that uncertainty by design.

Security leaders should prioritize controls before AI is fully deployed, avoid generic AI security checklists, integrate AI risk into existing cyber programs, and mitigate the risk of non-deterministic technology with continuous oversight, monitoring, behavior analytics, anomaly detection, autonomous investigation, and autonomous containment.

Visibility has a different connotation with AI. Previously, audit logging worked for software/people, but with Generative AI-based systems, interpretability and explainability is difficult to understand, you cannot "undo" what has been done, or see the logic or control a chain of events. This is why behavioral-based detections and containment becomes critical.

What capabilities should every AI security program include?

If an organization asked “what must be in place before scaling AI?”:

AI Risk board and approval workflow

IAM + PAM for all AI services and agents

AI asset inventory

Prompt/output DLP with sanctioned AI access – This is not just pre- and post- filters, but behavior-based detections of semantic interface as well as behavior-based analysis of output with associated risk context.

Shadow AI identification

Secure MLOps – This is an entire paper itself

Runtime guardrails and tool restrictions
- Including AI Gateway/SASE/Zero trust/

Runtime security with behavior-based detections
- Complete visibility, monitoring, behavior analytics, anomaly detection, risk/intent/context evaluation of anomalies, autonomous investigation and autonomous containment of all AI assets across endpoint, network, SaaS, SASE, cloud, OT, email, and messaging platforms

Secure data pipelines and data governance

SOC workflow changes from malicious classification workflows to behavior-based detection workflows

Remediation plans for AI-related incidents

Layered Governance and Security Stack for Securing AI

The following outline considers governance and security tools that should be considered, well-integrated, deployed, tested, operationalized and embedded within security workflows. These tools and controls map to NIST’s CMF for AI.

These considerations do not need to be implemented in order. Runtime Detect and Respond will help mitigate risk while Governance, Visibility, and Identity mature.

Category	Tooling	Controls
Governance & Visibility	AI asset inventory / AI CMDB Shadow AI discovery SaaS discovery AI usage on non-endpoint managed systems via network or cloud telemetry MCP server/client usage via protocols Browser telemetry Gateway or SASE telemetry	Establish a risk board to set up controls Mandatory registration of AI systems Owner, data classification, intended use, and risk tier Supplier disclosure requirements Risk mitigation plan for AI adoption, innovation, or development
Identity, Access & Agent Control	Non-human autonomous agents should not have the full permissions associated with a human user. IAM with workload identities PAM for AI service accounts Secrets management with short-lived tokens Zero Trust principles Identity, permission, and token hygiene	Unique identities per model, agent, and pipeline Least privilege for tools, data, and APIs Explicit approval for autonomous actions
Data Security & Privacy	Data classification and labeling Enterprise DLP across endpoint, email, network, cloud, and SaaS Forensics data capture after risky detections	Prompt-level DLP through behavior-based semantic analysis with risk and intent context Input/interface analysis for risky data requests Output analysis for sensitive data Data integrity evaluation Retention and redaction policies for prompts and responses
Secure MLOps / LLMOps	Secure CI/CD with AI-specific gates Model registries with approval workflows Dependency, container, and artifact scanning SBOM/AIBOM generation IaC security scanning Security posture management Misconfiguration identification Hardening recommendations	Signed models and prompts Versioned datasets, configurations, logging, and controls Securing data pipelines Controlled promotion Quality assurance Adversarial testing
Runtime Security	Securing runtime goes beyond guardrails and model firewalls to include behavior-based detections, response, and containment. Detection, monitoring, and SOC integration Centralized visibility into prompts, outputs, and tool calls AI-specific detections Behavior-based detection for AI usage patterns Model drift and behavior monitoring Autonomous containment Behavior-based detection of model inputs and outputs Prompt injection detection Model manipulation, including jailbreaking, poisoning, and related attacks Sensitive data access attempts Behavior-based detection across low-code agents, high-code agents, MCP clients and servers, endpoint, network, cloud, email, SaaS, SASE, IoT, and OT	Policy enforcement between users, models, tools, agents, SaaS models/tools, and MCP servers/clients Risk, intent, and context evaluation for detections and response actions
Response & Recovery	Autonomous containment AI-assisted playbooks Forensics data capture for AI-related events Model rollback mechanisms Backup and restore for models and datasets	Kill switch for agents Autonomous response to agents performing risky behaviors Model and dataset rollback Remediation plans Tabletop exercises Supplier coordination plans Post-incident AI performance validation

AI security requires continuous visibility and behavioral detection

AI changes how fast systems move, how decisions are made, and how risk propagates. It does not change the fundamentals of security. Organizations that succeed will be the ones that apply those fundamentals rigorously, assume failure, and build systems that can detect, contain, and recover when AI behaves in ways they did not anticipate. Security is not what AI is allowed to do. It is whether the organization can understand, trust, and control what AI actually does in practice.

Take this guidance to understand different initiatives that organizations should be considering. Securing AI is the most critical component to AI safety. As organizations invest more in AI adoption, they should be investing in security in order to mitigate the risk of AI adoption. Organizations should be evaluating their governance and security stack to include well-integrated tools that are deployed, tested, operationalized and embedded within security workflows. While organizations mature in governance, visibility and identity access management, they should be investing in behavior-based detection and autonomous containment to mitigate AI risk.

About the author

Blog

July 6, 2026

NIST Just Proved It: AI Security Can’t Be Solved With Rules

Static AI guardrails are inherently limited

As organizations adopt generative AI, many still assume that the right set of guardrails will be enough. The problem is you can’t anticipate every way these systems might be misused, abused or attacked. What NIST has done is put a mathematical foundation under that intuition.

In recent research building on Gödel’s incompleteness theorems, which showed that any system built on a fixed set of rules will always have gaps, NIST demonstrates that there is no finite set of guardrails that can be universally robust against adversarial prompts. In plain terms, if your defense is based on a fixed set of rules, there will always be inputs that bypass them. Not because the rules are badly written, but because the problem space is bigger than static rules can ever cover.

This is not new in cybersecurity - detection rules have always had to live with this trade-off. What is different with GenAI is the scale and shape of that problem. These systems are built on human language, and human language is not bounded. It is fluid, contextual and deliberately ambiguous. The number of ways intent can be hidden is effectively limitless. You are not defending against a defined protocol or a fixed exploit chain. You are defending against the entire expressive capacity of people.

So attempting to create a complete set of rules is the wrong starting point. It assumes the problem can be deterministically described. NIST’s work shows that it cannot. Organizations still need a way to manage AI risk, but the traditional approach of defining allowed and disallowed patterns is always going to lag behind what is actually happening. The same input can be benign in one context and risky in another, and static rules struggle to capture that distinction.

The question then is what fills that gap?

AI security must shift from rules to behavior

What's required is a shift in what you are trying to understand. Rules try to describe what should and shouldn't happen. Behavior shows you what is happening. Or to put it another way, if inputs are unbounded and adversaries adapt, the only stable signal is behavior.

In a GenAI context, that means analyzing how an AI model is being used, how prompts evolve over time, how outputs are shaped, and where AI agent interactions start to drift from what is expected. It means moving from static definitions of bad to a more dynamic understanding of intent.

Instead of trying to predict every bad prompt, you focus on identifying when behavior starts to move outside expected norms. Instead of asking whether a single input matches a rule, you ask whether the overall pattern of activity makes sense for the system and how it’s being used.

Guardrails remain important but they are only one layer

This does not eliminate the need for guardrails. They still play a role. But they will never address the entire problem space and are simply one part of your defense in depth approach.

NIST’s proof is useful because it makes this explicit. It removes the assumption that with enough effort, a complete rule set is achievable. It isn’t.

Once you accept that, the shift becomes unavoidable. This is no longer a problem of writing better rules, but of understanding behavior in a space where the possible inputs are effectively unbounded.

For security leaders, that changes the nature of the problem. It is less about defining what should be allowed, and more about recognizing when something is no longer consistent with expected behavior.

That does not remove the need for guardrails, but it does change their role. They set boundaries, but they do not define understanding. The gap between the two is where risk now sits.

In the end, this is what “can’t be solved with rules” really means. Rules will always leave gaps, and those gaps are not theoretical. They show up in how systems actually behave Not what we expect them to do, or what we intended them to do, but what they are doing in practice. That is where the signal is, and increasingly, that is where the security problem sits.

References:

https://www.nist.gov/news-events/news/2026/06/nist-mathematical-proof-supports-transition-continuous-monitor-and-update

https://ieeexplore.ieee.org/document/11475847

About the author

Andrew Hollister

Principal Solutions Engineer, Cyber Technician

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Check out this article by Darktrace: Hola VPN Abuse: From Proxy Traffic to Malware and Cryptomining

Hola VPN Abuse: From Proxy Traffic to Malware and Cryptomining

Introduction

The campaign

Case Study 1

Darktrace’s Autonomous Response

Case Study 2

Darktrace’s Autonomous Response

Suspected cryptomining activity

Conclusion

Appendices

References

Darktrace Model Detections

Indicators of Compromise (IoCs)

MITRE ATT&CK Mapping

From Click to Command: Behavioral Detection of AppleScript-Led MacOS Intrusions

Securing AI: Analysis of the Complete Security Stack with Governance and Controls

Enjoying the blog?

More in this series

From Click to Command: Behavioral Detection of AppleScript-Led MacOS Intrusions

How Attackers Abuse the Chinese Nezha Monitoring Tool

Darktrace named a Leader in the 2026 Gartner® Magic Quadrant™ for Network Detection and Response (NDR) For the Second Consecutive Year

Blog

July 7, 2026

Securing AI: Analysis of the Complete Security Stack with Governance and Controls

Why traditional cybersecurity approaches are not enough for AI

How NIST defines the three core pillars of AI security

Secure

Defend

Thwart

How this looks in practice

Why agentic AI fundamentally changes enterprise cyber risk

Behavioral risk

Structural risk

Accountability

What does defense in depth look like for AI?

How security leaders should prepare for enterprise AI adoption

What capabilities should every AI security program include?

Layered Governance and Security Stack for Securing AI

AI security requires continuous visibility and behavioral detection

Blog

July 6, 2026

NIST Just Proved It: AI Security Can’t Be Solved With Rules

Static AI guardrails are inherently limited

AI security must shift from rules to behavior

Guardrails remain important but they are only one layer