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
Deputy Team Lead, London & Cyber Analyst
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
Deputy Team Lead, London & Cyber Analyst

More in this series

No items found.

Blog

/

/

January 7, 2026

How a leading bank is prioritizing risk management to power a resilient future

Default blog imageDefault blog image

As one of the region’s most established financial institutions, this bank sits at the heart of its community’s economic life – powering everything from daily transactions to business growth and long-term wealth planning. Its blend of physical branches and advanced digital services gives customers the convenience they expect and the personal trust they rely on. But as the financial world becomes more interconnected and adversaries more sophisticated, safeguarding that trust requires more than traditional cybersecurity. It demands a resilient, forward-leaning approach that keeps pace with rising threats and tightening regulatory standards.

A complex risk landscape demands a new approach

The bank faced a challenge familiar across the financial sector: too many tools, not enough clarity. Vulnerability scans, pen tests, and risk reports all produced data, yet none worked together to show how exposures connected across systems or what they meant for day-to-day operations. Without a central platform to link and contextualize this data, teams struggled to see how individual findings translated into real exposure across the business.

  • Fragmented risk assessments: Cyber and operational risks were evaluated in silos, often duplicated across teams, and lacked the context needed to prioritize what truly mattered.
  • Limited executive visibility: Leadership struggled to gain a complete, real-time view of trends or progress, making risk ownership difficult to enforce.
  • Emerging compliance pressure: This gap also posed compliance challenges under the EU’s Digital Operational Resilience Act (DORA), which requires financial institutions to demonstrate continuous oversight, effective reporting, and the ability to withstand and recover from cyber and IT disruptions.
“The issue wasn’t the lack of data,” recalls the bank’s Chief Technology Officer. “The challenge was transforming that data into a unified, contextualized picture we could act on quickly and decisively.”

As the bank advanced its digital capabilities and embraced cloud services, its risk environment became more intricate. New pathways for exploitation emerged, human factors grew harder to quantify, and manual processes hindered timely decision-making. To maintain resilience, the security team sought a proactive, AI-powered platform that could consolidate exposures, deliver continuous insight, and ensure high-value risks were addressed before they escalated.

Choosing Darktrace to unlock proactive cyber resilience

To reclaim control over its fragmented risk landscape, the bank selected Darktrace / Proactive Exposure Management™ for cyber risk insight. The solution’s ability to consolidate scanner outputs, pen test results, CVE data, and operational context into one AI-powered view made it the clear choice. Darktrace delivered comprehensive visibility the team had long been missing.

By shifting from a reactive model to proactive security, the bank aimed to:

  • Improve resilience and compliance with DORA
  • Prioritize remediation efforts with greater accuracy
  • Eliminate duplicated work across teams
  • Provide leadership with a complete view of risk, updated continuously
  • Reduce the overall likelihood of attack or disruption

The CTO explains: “We needed a solution that didn’t just list vulnerabilities but showed us what mattered most for our business – how risks connected, how they could be exploited, and what actions would create the biggest reduction in exposure. Darktrace gave us that clarity.”

Targeting the risks that matter most

Darktrace / Proactive Exposure Management offered the bank a new level of visibility and control by continuously analyzing misconfigurations, critical attack paths, human communication patterns, and high-value assets. Its AI-driven risk scoring allowed the team to understand which vulnerabilities had meaningful business impact, not just which were technically severe.

Unifying exposure across architectures

Darktrace aggregates and contextualizes data from across the bank’s security stack, eliminating the need to manually compile or correlate findings. What once required hours of cross-team coordination now appears in a single, continuously updated dashboard.

Revealing an adversarial view of risk

The solution maps multi-stage, complex attack paths across network, cloud, identity systems, email environments, and endpoints – highlighting risks that traditional CVE lists overlook.

Identifying misconfigurations and controlling gaps

Using Self-Learning AI, Darktrace / Proactive Exposure Management spots misconfigurations and prioritizes them based on MITRE adversary techniques, business context, and the bank’s unique digital environment.

Enhancing red-team and pen test effectiveness

By directing testers to the highest-value targets, Darktrace removes guesswork and validates whether defenses hold up against realistic adversarial behavior.

Supporting DORA compliance

From continuous monitoring to executive-ready reporting, the solution provides the transparency and accountability the bank needs to demonstrate operational resilience frameworks.

Proactive security delivers tangible outcomes

Since deploying Darktrace / Proactive Exposure Management, the bank has significantly strengthened its cybersecurity posture while improving operational efficiency.

Greater insight, smarter prioritization, stronger defensee

Security teams are now saving more than four hours per week previously spent aggregating and analyzing risk data. With a unified view of their exposure, they can focus directly on remediation instead of manually correlating multiple reports.

Because risks are now prioritized based on business impact and real-time operational context, they no longer waste time on low-value tasks. Instead, critical issues are identified and resolved sooner, reducing potential windows for exploitation and strengthening the bank’s ongoing resilience against both known and emerging threats.

“Our goal was to move from reactive to proactive security,” the CTO says. “Darktrace didn’t just help us achieve that, it accelerated our roadmap. We now understand our environment with a level of clarity we simply didn’t have before.”

Leadership clarity and stronger governance

Executives and board stakeholders now receive clear, organization-wide visibility into the bank’s risk posture, supported by consistent reporting that highlights trends, progress, and areas requiring attention. This transparency has strengthened confidence in the bank’s cyber resilience and enabled leadership to take true ownership of risk across the institution.

Beyond improved visibility, the bank has also deepened its overall governance maturity. Continuous monitoring and structured oversight allow leaders to make faster, more informed decisions that strategically align security efforts with business priorities. With a more predictable understanding of exposure and risk movement over time, the organization can maintain operational continuity, demonstrate accountability, and adapt more effectively as regulatory expectations evolve.

Trading stress for control

With Darktrace, leaders now have the clarity and confidence they need to report to executives and regulators with accuracy. The ability to see organization-wide risk in context provides assurance that the right issues are being addressed at the right time. That clarity is also empowering security analysts who no longer shoulder the anxiety of wondering which risks matter most or whether something critical has slipped through the cracks. Instead, they’re working with focus and intention, redirecting hours of manual effort into strategic initiatives that strengthen the bank’s overall resilience.

Prioritizing risk to power a resilient future

For this leading financial institution, Darktrace / Proactive Exposure Management has become the foundation for a more unified, data-driven, and resilient cybersecurity program. With clearer, business-relevant priorities, stronger oversight, and measurable efficiency gains, the bank has strengthened its resilience and met demanding regulatory expectations without adding operational strain.

Most importantly, it shifted the bank’s security posture from a reactive stance to a proactive, continuous program. Giving teams the confidence and intelligence to anticipate threats and safeguard the people and services that depend on them.

Continue reading
About the author
Kelland Goodin
Product Marketing Specialist

Blog

/

/

December 22, 2025

The Year Ahead: AI Cybersecurity Trends to Watch in 2026

2026 cyber threat trendsDefault blog imageDefault blog image

Introduction: 2026 cyber trends

Each year, we ask some of our experts to step back from the day-to-day pace of incidents, vulnerabilities, and headlines to reflect on the forces reshaping the threat landscape. The goal is simple:  to identify and share the trends we believe will matter most in the year ahead, based on the real-world challenges our customers are facing, the technology and issues our R&D teams are exploring, and our observations of how both attackers and defenders are adapting.  

In 2025, we saw generative AI and early agentic systems moving from limited pilots into more widespread adoption across enterprises. Generative AI tools became embedded in SaaS products and enterprise workflows we rely on every day, AI agents gained more access to data and systems, and we saw glimpses of how threat actors can manipulate commercial AI models for attacks. At the same time, expanding cloud and SaaS ecosystems and the increasing use of automation continued to stretch traditional security assumptions.

Looking ahead to 2026, we’re already seeing the security of AI models, agents, and the identities that power them becoming a key point of tension – and opportunity -- for both attackers and defenders. Long-standing challenges and risks such as identity, trust, data integrity, and human decision-making will not disappear, but AI and automation will increase the speed and scale of the cyber risk.  

Here's what a few of our experts believe are the trends that will shape this next phase of cybersecurity, and the realities organizations should prepare for.  

Agentic AI is the next big insider risk

In 2026, organizations may experience their first large-scale security incidents driven by agentic AI behaving in unintended ways—not necessarily due to malicious intent, but because of how easily agents can be influenced. AI agents are designed to be helpful, lack judgment, and operate without understanding context or consequence. This makes them highly efficient—and highly pliable. Unlike human insiders, agentic systems do not need to be socially engineered, coerced, or bribed. They only need to be prompted creatively, misinterpret legitimate prompts, or be vulnerable to indirect prompt injection. Without strong controls around access, scope, and behavior, agents may over-share data, misroute communications, or take actions that introduce real business risk. Securing AI adoption will increasingly depend on treating agents as first-class identities—monitored, constrained, and evaluated based on behavior, not intent.

-- Nicole Carignan, SVP of Security & AI Strategy

Prompt Injection moves from theory to front-page breach

We’ll see the first major story of an indirect prompt injection attack against companies adopting AI either through an accessible chatbot or an agentic system ingesting a hidden prompt. In practice, this may result in unauthorized data exposure or unintended malicious behavior by AI systems, such as over-sharing information, misrouting communications, or acting outside their intended scope. Recent attention on this risk—particularly in the context of AI-powered browsers and additional safety layers being introduced to guide agent behavior—highlights a growing industry awareness of the challenge.  

-- Collin Chapleau, Senior Director of Security & AI Strategy

Humans are even more outpaced, but not broken

When it comes to cyber, people aren’t failing; the system is moving faster than they can. Attackers exploit the gap between human judgment and machine-speed operations. The rise of deepfakes and emotion-driven scams that we’ve seen in the last few years reduce our ability to spot the familiar human cues we’ve been taught to look out for. Fraud now spans social platforms, encrypted chat, and instant payments in minutes. Expecting humans to be the last line of defense is unrealistic.

Defense must assume human fallibility and design accordingly. Automated provenance checks, cryptographic signatures, and dual-channel verification should precede human judgment. Training still matters, but it cannot close the gap alone. In the year ahead, we need to see more of a focus on partnership: systems that absorb risk so humans make decisions in context, not under pressure.

-- Margaret Cunningham, VP of Security & AI Strategy

AI removes the attacker bottleneck—smaller organizations feel the impact

One factor that is currently preventing more companies from breaches is a bottleneck on the attacker side: there’s not enough human hacker capital. The number of human hands on a keyboard is a rate-determining factor in the threat landscape. Further advancements of AI and automation will continue to open that bottleneck. We are already seeing that. The ostrich approach of hoping that one’s own company is too obscure to be noticed by attackers will no longer work as attacker capacity increases.  

-- Max Heinemeyer, Global Field CISO

SaaS platforms become the preferred supply chain target

Attackers have learned a simple lesson: compromising SaaS platforms can have big payouts. As a result, we’ll see more targeting of commercial off-the-shelf SaaS providers, which are often highly trusted and deeply integrated into business environments. Some of these attacks may involve software with unfamiliar brand names, but their downstream impact will be significant. In 2026, expect more breaches where attackers leverage valid credentials, APIs, or misconfigurations to bypass traditional defenses entirely.

-- Nathaniel Jones, VP of Security & AI Strategy

Increased commercialization of generative AI and AI assistants in cyber attacks

One trend we’re watching closely for 2026 is the commercialization of AI-assisted cybercrime. For example, cybercrime prompt playbooks sold on the dark web—essentially copy-and-paste frameworks that show attackers how to misuse or jailbreak AI models. It’s an evolution of what we saw in 2025, where AI lowered the barrier to entry. In 2026, those techniques become productized, scalable, and much easier to reuse.  

-- Toby Lewis, Global Head of Threat Analysis

Conclusion

Taken together, these trends underscore that the core challenges of cybersecurity are not changing dramatically -- identity, trust, data, and human decision-making still sit at the core of most incidents. What is changing quickly is the environment in which these challenges play out. AI and automation are accelerating everything: how quickly attackers can scale, how widely risk is distributed, and how easily unintended behavior can create real impact. And as technology like cloud services and SaaS platforms become even more deeply integrated into businesses, the potential attack surface continues to expand.  

Predictions are not guarantees. But the patterns emerging today suggest that 2026 will be a year where securing AI becomes inseparable from securing the business itself. The organizations that prepare now—by understanding how AI is used, how it behaves, and how it can be misused—will be best positioned to adopt these technologies with confidence in the year ahead.

Learn more about how to secure AI adoption in the enterprise without compromise by registering to join our live launch webinar on February 3, 2026.  

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