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!

Written by

Charlotte Thompson

Cyber Analyst

Written by

Tiana Kelly

Deputy Team Lead, London & Cyber 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

Charlotte Thompson

Cyber Analyst

Written by

Tiana Kelly

Deputy Team Lead, London & Cyber Analyst

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/

Deputy Team Lead, London & Cyber Analyst

Inside the SOC

Written by

Charlotte Thompson

Cyber Analyst

Written by

Tiana Kelly

Deputy Team Lead, London & Cyber Analyst

Senior Product Marketing Manager, Email

Watch the NIS2 Webinar

Blog

Identity

July 3, 2025

Top Eight Threats to SaaS Security and How to Combat Them

The latest on the identity security landscape

Following the mass adoption of remote and hybrid working patterns, more critical data than ever resides in cloud applications – from Salesforce and Google Workspace, to Box, Dropbox, and Microsoft 365.

On average, a single organization uses 130 different Software-as-a-Service (SaaS) applications, and 45% of organizations reported experiencing a cybersecurity incident through a SaaS application in the last year.

As SaaS applications look set to remain an integral part of the digital estate, organizations are being forced to rethink how they protect their users and data in this area.

What is SaaS security?

SaaS security is the protection of cloud applications. It includes securing the apps themselves as well as the user identities that engage with them.

Below are the top eight threats that target SaaS security and user identities.

1. Account Takeover (ATO)

Attackers gain unauthorized access to a user’s SaaS or cloud account by stealing credentials through phishing, brute-force attacks, or credential stuffing. Once inside, they can exfiltrate data, send malicious emails, or escalate privileges to maintain persistent access.

2. Privilege escalation

Cybercriminals exploit misconfigurations, weak access controls, or vulnerabilities to increase their access privileges within a SaaS or cloud environment. Gaining admin or superuser rights allows attackers to disable security settings, create new accounts, or move laterally across the organization.

3. Lateral movement

Once inside a network or SaaS platform, attackers move between accounts, applications, and cloud workloads to expand their foot- hold. Compromised OAuth tokens, session hijacking, or exploited API connections can enable adversaries to escalate access and exfiltrate sensitive data.

4. Multi-Factor Authentication (MFA) bypass and session hijacking

Threat actors bypass MFA through SIM swapping, push bombing, or exploiting session cookies. By stealing an active authentication session, they can access SaaS environments without needing the original credentials or MFA approval.

5. OAuth token abuse

Attackers exploit OAuth authentication mechanisms by stealing or abusing tokens that grant persistent access to SaaS applications. This allows them to maintain access even if the original user resets their password, making detection and mitigation difficult.

6. Insider threats

Malicious or negligent insiders misuse their legitimate access to SaaS applications or cloud platforms to leak data, alter configurations, or assist external attackers. Over-provisioned accounts and poor access control policies make it easier for insiders to exploit SaaS environments.

7. Application Programming Interface (API)-based attacks

SaaS applications rely on APIs for integration and automation, but attackers exploit insecure endpoints, excessive permissions, and unmonitored API calls to gain unauthorized access. API abuse can lead to data exfiltration, privilege escalation, and service disruption.

8. Business Email Compromise (BEC) via SaaS

Adversaries compromise SaaS-based email platforms (e.g., Microsoft 365 and Google Workspace) to send phishing emails, conduct invoice fraud, or steal sensitive communications. BEC attacks often involve financial fraud or data theft by impersonating executives or suppliers.

BEC heavily uses social engineering techniques, tailoring messages for a specific audience and context. And with the growing use of generative AI by threat actors, BEC is becoming even harder to detect. By adding ingenuity and machine speed, generative AI tools give threat actors the ability to create more personalized, targeted, and convincing attacks at scale.

Protecting against these SaaS threats

Traditionally, security leaders relied on tools that were focused on the attack, reliant on threat intelligence, and confined to a single area of the digital estate.

However, these tools have limitations, and often prove inadequate for contemporary situations, environments, and threats. For example, they may lack advanced threat detection, have limited visibility and scope, and struggle to integrate with other tools and infrastructure, especially cloud platforms.

AI-powered SaaS security stays ahead of the threat landscape

New, more effective approaches involve AI-powered defense solutions that understand the digital business, reveal subtle deviations that indicate cyber-threats, and action autonomous, targeted responses.

[related-resource]

‍

About the author

Carlos Gray

Senior Product Marketing Manager, Email

Blog

July 2, 2025

Pre-CVE Threat Detection: 10 Examples Identifying Malicious Activity Prior to Public Disclosure of a Vulnerability

‍

Vulnerabilities are weaknesses in a system that can be exploited by malicious actors to gain unauthorized access or to disrupt normal operations. Common Vulnerabilities and Exposures (or CVEs) are a list of publicly disclosed cybersecurity vulnerabilities that can be tracked and mitigated by the security community.

When a vulnerability is discovered, the standard practice is to report it to the vendor or the responsible organization, allowing them to develop and distribute a patch or fix before the details are made public. This is known as responsible disclosure.

With a record-breaking 40,000 CVEs reported for 2024 and a predicted higher number for 2025 by the Forum for Incident Response and Security Teams (FIRST) [1], anomaly-detection is essential for identifying these potential risks. The gap between exploitation of a zero-day and disclosure of the vulnerability can sometimes be considerable, and retroactively attempting to identify successful exploitation on your network can be challenging, particularly if taking a signature-based approach.

Detecting threats without relying on CVE disclosure

Abnormal behaviors in networks or systems, such as unusual login patterns or data transfers, can indicate attempted cyber-attacks, insider threats, or compromised systems. Since Darktrace does not rely on rules or signatures, it can detect malicious activity that is anomalous even without full context of the specific device or asset in question.

‍For example, during the Fortinet exploitation late last year, the Darktrace Threat Research team were investigating a different Fortinet vulnerability, namely CVE 2024-23113, for exploitation when Mandiant released a security advisory around CVE 2024-47575, which aligned closely with Darktrace’s findings.

Retrospective analysis like this is used by Darktrace’s threat researchers to better understand detections across the threat landscape and to add additional context.

Below are ten examples from the past year where Darktrace detected malicious activity days or even weeks before a vulnerability was publicly disclosed.

ten examples from the past year where Darktrace detected malicious activity days or even weeks before a vulnerability was publicly disclosed.

Trends in pre-cve exploitation

Often, the disclosure of an exploited vulnerability can be off the back of an incident response investigation related to a compromise by an advanced threat actor using a zero-day. Once the vulnerability is registered and publicly disclosed as having been exploited, it can kick off a race between the attacker and defender: attack vs patch.

Nation-state actors, highly skilled with significant resources, are known to use a range of capabilities to achieve their target, including zero-day use. Often, pre-CVE activity is “low and slow”, last for months with high operational security. After CVE disclosure, the barriers to entry lower, allowing less skilled and less resourced attackers, like some ransomware gangs, to exploit the vulnerability and cause harm. This is why two distinct types of activity are often seen: pre and post disclosure of an exploited vulnerability.

Darktrace saw this consistent story line play out during several of the Fortinet and PAN OS threat actor campaigns highlighted above last year, where nation-state actors were seen exploiting vulnerabilities first, followed by ransomware gangs impacting organizations [2].

The same applies with the recent SAP Netweaver exploitations being tied to a China based threat actor earlier this spring with subsequent ransomware incidents being observed [3].

Autonomous Response

Anomaly-based detection offers the benefit of identifying malicious activity even before a CVE is disclosed; however, security teams still need to quickly contain and isolate the activity.

For example, during the Ivanti chaining exploitation in the early part of 2025, a customer had Darktrace’s Autonomous Response capability enabled on their network. As a result, Darktrace was able to contain the compromise and shut down any ongoing suspicious connectivity by blocking internal connections and enforcing a “pattern of life” on the affected device.

This pre-CVE detection and response by Darktrace occurred 11 days before any public disclosure, demonstrating the value of an anomaly-based approach.

In some cases, customers have even reported that Darktrace stopped malicious exploitation of devices several days before a public disclosure of a vulnerability.

For example, During the ConnectWise exploitation, a customer informed the team that Darktrace had detected malicious software being installed via remote access. Upon further investigation, four servers were found to be impacted, while Autonomous Response had blocked outbound connections and enforced patterns of life on impacted devices.

Conclusion

By continuously analyzing behavioral patterns, systems can spot unusual activities and patterns from users, systems, and networks to detect anomalies that could signify a security breach.

Through ongoing monitoring and learning from these behaviors, anomaly-based security systems can detect threats that traditional signature-based solutions might miss, while also providing detailed insights into threat tactics, techniques, and procedures (TTPs). This type of behavioral intelligence supports pre-CVE detection, allows for a more adaptive security posture, and enables systems to evolve with the ever-changing threat landscape.

Credit to Nathaniel Jones (VP, Security & AI Strategy, Field CISO), Emma Fougler (Global Threat Research Operations Lead), Ryan Traill (Analyst Content Lead)

References and further reading:

Related Darktrace blogs:

‍

*Self-reported by customer, confirmed afterwards.

**Updated January 2024 blog now reflects current findings

About the author

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Check out this article by Darktrace: Exploring AI Threats: Package Hallucination Attacks

Exploring AI Threats: Package Hallucination Attacks

AI tools open doors for threat actors

What is an AI hallucination?

Why are AI hallucinations a problem?

Use of generative AI in software development

Malicious packages as attack vectors

How could Darktrace detect an AI Package Hallucination Attack?

Conclusion

References

Pre-CVE Threat Detection: 10 Examples Identifying Malicious Activity Prior to Public Disclosure of a Vulnerability

Top Eight Threats to SaaS Security and How to Combat Them

Enjoying the blog?

More in this series

Blog

Identity

July 3, 2025

Top Eight Threats to SaaS Security and How to Combat Them

The latest on the identity security landscape

What is SaaS security?

1. Account Takeover (ATO)

2. Privilege escalation

3. Lateral movement

4. Multi-Factor Authentication (MFA) bypass and session hijacking

5. OAuth token abuse

6. Insider threats

7. Application Programming Interface (API)-based attacks

8. Business Email Compromise (BEC) via SaaS

Protecting against these SaaS threats

AI-powered SaaS security stays ahead of the threat landscape

Blog

July 2, 2025

Pre-CVE Threat Detection: 10 Examples Identifying Malicious Activity Prior to Public Disclosure of a Vulnerability

Detecting threats without relying on CVE disclosure

Below are ten examples from the past year where Darktrace detected malicious activity days or even weeks before a vulnerability was publicly disclosed.

Trends in pre-cve exploitation

Autonomous Response

Conclusion

References and further reading:

Related Darktrace blogs: