Blog
/
/
October 30, 2023

Exploring AI Threats: Package Hallucination Attacks

Learn how malicious actors exploit errors in generative AI tools to launch packet attacks. Read how Darktrace products detect and prevent these threats!
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
Charlotte Thompson
Cyber Analyst
Written by
Tiana Kelly
Senior Cyber Analyst & Team Lead
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog image
30
Oct 2023

AI tools open doors for threat actors

On November 30, 2022, the free conversational language generation model ChatGPT was launched by OpenAI, an artificial intelligence (AI) research and development company. The launch of ChatGPT was the culmination of development ongoing since 2018 and represented the latest innovation in the ongoing generative AI boom and made the use of generative AI tools accessible to the general population for the first time.

ChatGPT is estimated to currently have at least 100 million users, and in August 2023 the site reached 1.43 billion visits [1]. Darktrace data indicated that, as of March 2023, 74% of active customer environments have employees using generative AI tools in the workplace [2].

However, with new tools come new opportunities for threat actors to exploit and use them maliciously, expanding their arsenal.

Much consideration has been given to mitigating the impacts of the increased linguistic complexity in social engineering and phishing attacks resulting from generative AI tool use, with Darktrace observing a 135% increase in ‘novel social engineering attacks’ across thousands of active Darktrace/Email™ customers from January to February 2023, corresponding with the widespread adoption of ChatGPT and its peers [3].

Less overall consideration, however, has been given to impacts stemming from errors intrinsic to generative AI tools. One of these errors is AI hallucinations.

What is an AI hallucination?

AI “hallucination” is a term which refers to the predictive elements of generative AI and LLMs’ AI model gives an unexpected or factually incorrect response which does not align with its machine learning training data [4]. This differs from regular and intended behavior for an AI model, which should provide a response based on the data it was trained upon.  

Why are AI hallucinations a problem?

Despite the term indicating it might be a rare phenomenon, hallucinations are far more likely than accurate or factual results as the AI models used in LLMs are merely predictive and focus on the most probable text or outcome, rather than factual accuracy.

Given the widespread use of generative AI tools in the workplace employees are becoming significantly more likely to encounter an AI hallucination. Furthermore, if these fabricated hallucination responses are taken at face value, they could cause significant issues for an organization.

Use of generative AI in software development

Software developers may use generative AI for recommendations on how to optimize their scripts or code, or to find packages to import into their code for various uses. Software developers may ask LLMs for recommendations on specific pieces of code or how to solve a specific problem, which will likely lead to a third-party package. It is possible that packages recommended by generative AI tools could represent AI hallucinations and the packages may not have been published, or, more accurately, the packages may not have been published prior to the date at which the training data for the model halts. If these hallucinations result in common suggestions of a non-existent package, and the developer copies the code snippet wholesale, this may leave the exchanges vulnerable to attack.

Research conducted by Vulcan revealed the prevalence of AI hallucinations when ChatGPT is asked questions related to coding. After sourcing a sample of commonly asked coding questions from Stack Overflow, a question-and-answer website for programmers, researchers queried ChatGPT (in the context of Node.js and Python) and reviewed its responses. In 20% of the responses provided by ChatGPT pertaining to Node.js at least one un-published package was included, whilst the figure sat at around 35% for Python [4].

Hallucinations can be unpredictable, but would-be attackers are able to find packages to create by asking generative AI tools generic questions and checking whether the suggested packages exist already. As such, attacks using this vector are unlikely to target specific organizations, instead posing more of a widespread threat to users of generative AI tools.

Malicious packages as attack vectors

Although AI hallucinations can be unpredictable, and responses given by generative AI tools may not always be consistent, malicious actors are able to discover AI hallucinations by adopting the approach used by Vulcan. This allows hallucinated packages to be used as attack vectors. Once a malicious actor has discovered a hallucination of an un-published package, they are able to create a package with the same name and include a malicious payload, before publishing it. This is known as a malicious package.

Malicious packages could also be recommended by generative AI tools in the form of pre-existing packages. A user may be recommended a package that had previously been confirmed to contain malicious content, or a package that is no longer maintained and, therefore, is more vulnerable to hijack by malicious actors.

In such scenarios it is not necessary to manipulate the training data (data poisoning) to achieve the desired outcome for the malicious actor, thus a complex and time-consuming attack phase can easily be bypassed.

An unsuspecting software developer may incorporate a malicious package into their code, rendering it harmful. Deployment of this code could then result in compromise and escalation into a full-blown cyber-attack.

Figure 1: Flow diagram depicting the initial stages of an AI Package Hallucination Attack.

For providers of Software-as-a-Service (SaaS) products, this attack vector may represent an even greater risk. Such organizations may have a higher proportion of employed software developers than other organizations of comparable size. A threat actor, therefore, could utilize this attack vector as part of a supply chain attack, whereby a malicious payload becomes incorporated into trusted software and is then distributed to multiple customers. This type of attack could have severe consequences including data loss, the downtime of critical systems, and reputational damage.

How could Darktrace detect an AI Package Hallucination Attack?

In June 2023, Darktrace introduced a range of DETECT™ and RESPOND™ models designed to identify the use of generative AI tools within customer environments, and to autonomously perform inhibitive actions in response to such detections. These models will trigger based on connections to endpoints associated with generative AI tools, as such, Darktrace’s detection of an AI Package Hallucination Attack would likely begin with the breaching of one of the following DETECT models:

  • Compliance / Anomalous Upload to Generative AI
  • Compliance / Beaconing to Rare Generative AI and Generative AI
  • Compliance / Generative AI

Should generative AI tool use not be permitted by an organization, the Darktrace RESPOND model ‘Antigena / Network / Compliance / Antigena Generative AI Block’ can be activated to autonomously block connections to endpoints associated with generative AI, thus preventing an AI Package Hallucination attack before it can take hold.

Once a malicious package has been recommended, it may be downloaded from GitHub, a platform and cloud-based service used to store and manage code. Darktrace DETECT is able to identify when a device has performed a download from an open-source repository such as GitHub using the following models:

  • Device / Anomalous GitHub Download
  • Device / Anomalous Script Download Followed By Additional Packages

Whatever goal the malicious package has been designed to fulfil will determine the next stages of the attack. Due to their highly flexible nature, AI package hallucinations could be used as an attack vector to deliver a large variety of different malware types.

As GitHub is a commonly used service by software developers and IT professionals alike, traditional security tools may not alert customer security teams to such GitHub downloads, meaning malicious downloads may go undetected. Darktrace’s anomaly-based approach to threat detection, however, enables it to recognize subtle deviations in a device’s pre-established pattern of life which may be indicative of an emerging attack.

Subsequent anomalous activity representing the possible progression of the kill chain as part of an AI Package Hallucination Attack could then trigger an Enhanced Monitoring model. Enhanced Monitoring models are high-fidelity indicators of potential malicious activity that are investigated by the Darktrace analyst team as part of the Proactive Threat Notification (PTN) service offered by the Darktrace Security Operation Center (SOC).

Conclusion

Employees are often considered the first line of defense in cyber security; this is particularly true in the face of an AI Package Hallucination Attack.

As the use of generative AI becomes more accessible and an increasingly prevalent tool in an attacker’s toolbox, organizations will benefit from implementing company-wide policies to define expectations surrounding the use of such tools. It is simple, yet critical, for example, for employees to fact check responses provided to them by generative AI tools. All packages recommended by generative AI should also be checked by reviewing non-generated data from either external third-party or internal sources. It is also good practice to adopt caution when downloading packages with very few downloads as it could indicate the package is untrustworthy or malicious.

As of September 2023, ChatGPT Plus and Enterprise users were able to use the tool to browse the internet, expanding the data ChatGPT can access beyond the previous training data cut-off of September 2021 [5]. This feature will be expanded to all users soon [6]. ChatGPT providing up-to-date responses could prompt the evolution of this attack vector, allowing attackers to publish malicious packages which could subsequently be recommended by ChatGPT.

It is inevitable that a greater embrace of AI tools in the workplace will be seen in the coming years as the AI technology advances and existing tools become less novel and more familiar. By fighting fire with fire, using AI technology to identify AI usage, Darktrace is uniquely placed to detect and take preventative action against malicious actors capitalizing on the AI boom.

Credit to Charlotte Thompson, Cyber Analyst, Tiana Kelly, Analyst Team Lead, London, Cyber Analyst

References

[1] https://seo.ai/blog/chatgpt-user-statistics-facts

[2] https://darktrace.com/news/darktrace-addresses-generative-ai-concerns

[3] https://darktrace.com/news/darktrace-email-defends-organizations-against-evolving-cyber-threat-landscape

[4] https://vulcan.io/blog/ai-hallucinations-package-risk?nab=1&utm_referrer=https%3A%2F%2Fwww.google.com%2F

[5] https://twitter.com/OpenAI/status/1707077710047216095

[6] https://www.reuters.com/technology/openai-says-chatgpt-can-now-browse-internet-2023-09-27/

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
Charlotte Thompson
Cyber Analyst
Written by
Tiana Kelly
Senior Cyber Analyst & Team Lead

More in this series

No items found.

Blog

/

Network

/

April 2, 2026

How Chinese-Nexus Cyber Operations Have Evolved – And What It Means For Cyber Risk and Resilience 

Chinese-Nexus Cyber OperationsDefault blog imageDefault blog image

Cybersecurity has traditionally organized risk around incidents, breaches, campaigns, and threat groups. Those elements still matter—but if we fixate on individual incidents, we risk missing the shaping of the entire ecosystem. Nation‑state–aligned operators are increasingly using cyber operations to establish long-term strategic leverage, not just to execute isolated attacks or short‑term objectives.  

Our latest research, Crimson Echo, shifts the lens accordingly. Instead of dissecting campaigns, malware families, or actor labels as discrete events, the threat research team analyzed Chinese‑nexus activity as a continuum of behaviors over time. That broader view reveals how these operators position themselves within environments: quietly, patiently, and persistently—often preparing the ground long before any recognizable “incident” occurs.  

How Chinese-nexus cyber threats have changed over time

Chinese-nexus cyber activity has evolved in four phases over the past two decades. This ranges from early, high-volume operations in the 1990s and early 2000s to more structured, strategically-aligned activity in the 2010s, and now toward highly adaptive, identity-centric intrusions.  

Today’s phase is defined by scale, operational restraint, and persistence. Attackers are establishing access, evaluating its strategic value, and maintaining it over time. This reflects a broader shift: cyber operations are increasingly integrated into long-term economic and geopolitical strategies. Access to digital environments, specifically those tied to critical national infrastructure, supply chains, and advanced technology, has become a form of strategic leverage for the long-term.  

How Darktrace analysts took a behavioral approach to a complex problem

One of the challenges in analyzing nation-state cyber activity is attribution. Traditional approaches often rely on tracking specific threat groups, malware families, or infrastructure. But these change constantly, and in the case of Chinese-nexus operations, they often overlap.

Crimson Echo is the result of a retrospective analysis of three years of anomalous activity observed across the Darktrace fleet between July 2022 and September 2025. Using behavioral detection, threat hunting, open-source intelligence, and a structured attribution framework (the Darktrace Cybersecurity Attribution Framework), the team identified dozens of medium- to high-confidence cases and analyzed them for recurring operational patterns.  

This long-horizon, behavior-centric approach allows Darktrace to identify consistent patterns in how intrusions unfold, reinforcing that behavioral patterns that matter.  

What the data shows

Several clear trends emerged from the analysis:

  • Targeting is concentrated in strategically important sectors. Across the dataset, 88% of intrusions occurred in organizations classified as critical infrastructure, including transportation, critical manufacturing, telecommunications, government, healthcare, and Information Technology (IT) services.  
  • Strategically important Western economies are a primary focus. The US alone accounted for 22.5% of observed cases, and when combined with major European economies including Germany, Italy, Spain and the UK, over half of all intrusions (55%) were concentrated in these regions.  
  • Nearly 63% of intrusions of intrusions began with the exploitation of internet-facing systems, reinforcing the continued risk posed by externally exposed infrastructure.  

Two models of cyber operations

Across the dataset, Chinese-nexus activity followed two operational models.  

The first is best described as “smash and grab.” These are short-horizon intrusions optimized for speed. Attackers move quickly – often exfiltrating data within 48 hours – and prioritize scale over stealth. The median duration of these compromises is around 10 days. It’s clear they are willing to risk detection for short-term gain.  

The second is “low and slow.” These operations were less prevalent in the dataset, but potentially more consequential. Here, attackers prioritize persistence, establishing durable access through identity systems and legitimate administrative tools, so they can maintain access undetected for months or even years. In one notable case, the actor had fully compromised the environment and established persistence, only to resurface in the environment more than 600 days after. The operational pause underscores both the depth of the intrusion and the actor’s long‑term strategic intent. This suggests that cyber access is a strategic asset to preserve and leverage over time, and we observed these attacks most often inin sectors of the high strategic importance.  

It’s important to note that the same operational ecosystem can employ both models concurrently, selecting the appropriate model based on target value, urgency, intended access. The observation of a “smash and grab” model should not be solely interpreted as a failure of tradecraft, but instead an operational choice likely aligned with objectives. Where “low and slow” operations are optimized for patience, smash and grab is optimized for speed; both seemingly are deliberate operational choices, not necessarily indicators of capability.  

Rethinking cyber risk

For many organizations, cyber risk is still framed as a series of discrete events. Something happens, it is detected and contained, and the organization moves on. But persistent access, particularly in deeply interconnected environments that span cloud, identity-based SaaS and agentic systems, and complex supply chain networks, creates a major ongoing exposure risk. Even in the absence of disruption or data theft, that access can provide insight into operations, dependencies, and strategic decision-making. Cyber risk increasingly resembles long-term competitive intelligence.  

This has impact beyond the Security Operations Center. Organizations need to shift how they think about governance, visibility, and resilience, and treat cyber exposure as a structural business risk instead of an incident response challenge.  

What comes next

The goal of this research is to provide a clearer understanding of how these operations work, so defenders can recognize them earlier and respond more effectively. That includes shifting from tracking indicators to understanding behaviors, treating identity providers as critical infrastructure risks, expanding supplier oversight, investing in rapid containment capabilities, and more.  

Learn more about the findings of Darktrace’s latest research, Crimson Echo: Understanding Chinese-nexus Cyber Operations Through Behavioral Analysis, by downloading the full report and summaries for business leaders, CISOs, and SOC analysts here.  

Continue reading
About the author
Nathaniel Jones
VP, Security & AI Strategy, Field CISO

Blog

/

Proactive Security

/

April 1, 2026

AI-powered security for a rapidly growing grocery enterprise

Default blog imageDefault blog image

Protecting a complex, fast-growing retail organization

For this multi-banner grocery holding organization, cybersecurity is considered an essential business enabler, protecting operations, growth, and customer trust. The organization’s lean IT team manages a highly distributed environment spanning corporate offices, 100+ stores, distribution centers and  thousands of endpoints, users, and third-party connections.

Mergers and acquisitions fueled rapid growth, but they also introduced escalating complexity that constrained visibility into users, endpoints, and security risks inherited across acquired environments.

Closing critical visibility gaps with limited resources

Enterprise-wide visibility is a top priority for the organization, says the  Vice President of Information Technology. “We needed insights beyond the perimeter into how users and devices were behaving across the organization.”

A security breach that occurred before the current IT leadership joined the company reinforced the urgency and elevated cybersecurity to an executive-level priority with a focus on protecting customer trust. The goal was to build a multi-layered security model that could deliver autonomous, enterprise-wide protection without adding headcount.

Managing cyber risk in M&A

Mergers and acquisitions are central to the grocery holding company’s growth strategy. But each transaction introduces new cyber risk, including inherited network architectures, inconsistent tooling, excessive privileges, and remnants of prior security incidents that were never fully remediated.

“Our M&A targets range from small chains with a single IT person and limited cyber tools to large chains with more developed IT teams, toolsets and instrumentation,” explains the VP of IT. “We needed a fast, repeatable, and reliable way to assess cyber risk before transactions closed.”

AI-driven security built for scale, speed, and resilience

Rather than layering additional point tools onto an already complex environment, the retailer adopted the Darktrace ActiveAI Security Platform™ in 2020 as part of a broader modernization effort to improve resilience, close visibility gaps, and establish a security foundation that could scale with growth.

“Darktrace’s AI-driven approach provided the ideal solution to these challenges,” shares the VP of IT. “It has empowered our organization to maintain a robust security strategy, ensuring the protection of our network and the smooth operation of our business.”

Enterprise-wide visibility into traffic  

By monitoring both north-south and east-west traffic and applying Self-Learning AI, Darktrace develops a dynamic understanding of how users and devices normally behave across locations, roles, and systems.

“Modeling normal behavior across the environment enables us to quickly spot behavior that doesn’t fit. Even subtle changes that could signal a threat but appear legitimate at first glance,” explains the VP of IT.

Real-time threat containment, 24/7

Adopting autonomous response has created operational breathing room for the security team, says the company’s Cybersecurity  Engineer.

“Early on, we enabled full Darktrace autonomous mode and we continue to do so today,” shares the IT Security Architect. “Allowing the technology to act first gives us the time we need to investigate incidents during business hours without putting the business at risk.”

Unified, actionable view of security ecosystem

The grocery retailer integrated Darktrace with its existing security ecosystem of firewalls, vulnerability management tools, and endpoint detection and response, and the VP of IT described the adoption process as “exceptionally smooth.”

The team can correlate enterprise-wide security data for a unified and actionable picture of all activity and risk. Using this “single pane of glass” approach, the retailer trains Level 1 and Level 2 operations staff to assist with investigations and user follow-ups, effectively extending the reach of the security function without expanding headcount.

From reactive defense to security at scale

With Darktrace delivering continuous visibility, autonomous containment, and integrated security workflows, the organization has strengthened its cybersecurity posture while improving operational efficiency. The result is a security model that not only reduces risk, but also supports growth, resilience, and informed decision-making at the business level.

Faster detection, faster resolution

With autonomous detection and response, the retailer can immediately contain risk while analysts investigate and validate activity. With this approach, the company can maintain continuous protection even outside business hours and reduce the chance of lateral spread across systems or locations.

Enterprise-grade protection with a lean team

From cloud environments to clients to SaaS collaboration tools, Darktrace provides holistic autonomous AI defense, processing petabytes of the organization’s network traffic and investigating millions of individual events that could be indicative of a wider incident.

Today, Darktrace autonomously conducts the majority of all investigations on behalf of the IT team, escalating only a tiny fraction for analyst review. The impact has been profound, freeing analysts from endless alerts and hours of triage so they can focus on more valuable, proactive, and gratifying work.

“From an operational perspective, Darktrace gives us time back,” says the Cybersecurity Engineer. More importantly, says the VP of IT, “it gives us peace of mind that we’re protected even if we’re not actively monitoring every alert.”

A strategic input for M&A decision-making

One of the most strategic outcomes has been the role of cybersecurity on M&A. 90 days prior to closing a transaction, the security team uses Darktrace alongside other tools to perform a cyber risk assessment of the potential acquisition. “Our approach with Darktrace has consistently identified gaps and exposed risks,” says the VP of IT, including:

  • Remnants of previous incidents that were never fully remediated
  • Network configurations with direct internet exposure
  • Excessive administrative privileges in Active Directory or on critical hosts

While security findings may not alter deal timelines, the VP of IT says they can have enormous business implications. “With early visibility into these risks, we can reduce exposure to inherited cyber threats, strengthen our position during negotiations, and establish clear remediation requirements.”

A security strategy built to evolve with the business

As the holding group expands its cloud footprint, it will extend Darktrace protections into Azure, applying the same AI-driven visibility and autonomous response to cloud workloads. The VP of IT says Darktrace's evolving capabilities will be instrumental in addressing the organization’s future cybersecurity needs and ability to adapt to the dynamic nature of cloud security.

“With Darktrace’s AI-driven approach, we have moved beyond reactive defense, establishing a resilient security foundation for confident expansion and modernization.”

Continue reading
About the author
Your data. Our AI.
Elevate your network security with Darktrace AI