Blog
/
/
November 21, 2018

How Black Hats Take Advantage of Black Friday

The retail industry must be willing to adapt its cyber defenses against an ever-evolving adversary, or it may end Black Friday firmly in the red.
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
Justin Fier
SVP, Red Team Operations
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog image
21
Nov 2018

From Thanksgiving to Cyber Monday, shoppers across the globe will splurge tens of billions of dollars on everything from pillows to parkas to Pokémon pajamas.

U.S. consumers alone spent a record $19.62 billion last Black Friday weekend — on just online purchases. And while the number of customers at brick-and-mortar stores declined 4% from 2016, e-commerce sales were 18% higher in 2017, when for the first time more Americans shopped online than in person. There is every reason to suspect that a virtually unprecedented volume of virtual cash is about to change hands, presenting an equally unprecedented opportunity for a massive holiday cyber-heist. Here’s what such a heist might look like:

Proof of concept

While the incentive for cyber-crime during this Black Friday weekend is historically unparalleled, it has long been the holiday of choice for criminals. On Cyber Monday of 2014, for instance, a DNS provider was hit by a relatively rudimentary DDoS attack that nonetheless disrupted its clients’ websites. More advanced DDoS attacks launched by modern Mirai botnets — like the 2016 Dyn attack that crippled many of the Internet’s top websites — would be devastating on Black Friday, when companies like Amazon reel in upwards of a million dollars per minute. And for smaller retailers, a ransomware or DDoS attack this weekend poses existential risk, both because of lost revenue and because of reputational damage in such a highly competitive industry.

Prior to last year’s Black Friday weekend, experts anticipated more than 50 million attacks on businesses during peak shopping days, and cyber-criminals did not disappoint. Darktrace detected a 70% uptick in significant threats facing its retail clients during the holiday season, from November and December, compared to the previous two months, an uptick that helps explain why cyber-crime cost the world $600 billion last year. At least in the short term, it appears that online crime does pay — especially after Thanksgiving.

Mode of attack

As forensics continue to improve and CCTVs rapidly proliferate, the in-person criminal heist has largely been replaced by online robbery, which leaves no fingerprints and can be seen by no camera. One example: the annual amount of money stolen in U.S. bank robberies — the quintessential heist — has fallen by more than 60% since 2003, while cyber-crimes like credit card fraud have simultaneously skyrocketed. This transition to digital larceny makes financial sense as well, given that less than 10% of the world’s currency still exists as physical cash.

Indeed, identity theft is even more lucrative than bank robbery if done at scale, yet it entails far less risk for the perpetrators. Stolen credit card numbers can each sell for $100 on the Dark Web, rendering crimes like the Target breach — which took place during Black Friday weekend in 2013 and exposed 40 million debit and credit accounts — extremely profitable. With more than 100 million Americans and close to a billion global shoppers online during the holiday season, ’tis certainly the season for a large-scale assault on personal information.

But perhaps the most revolutionary aspect of cyber-heists is that they need not even steal anything to make off with loot. Faced with a well-timed ransomware attack, retailers often simply hand over their cash to remain operational: 70% of businesses paid the ransom after attacks in 2016, prompting criminals to quadruple their average demand. And on the busiest shopping day in history, there’s no telling how exorbitant these demands might be.

Cyber-threats that are specifically aimed at the retail sector make the challenge of security even more difficult for defenders, since much like a targeted traditional heist, they exploit their victims’ unique vulnerabilities. The numbers validate common sense here: insights from across Darktrace’s customer base reveal that these key retail threats — which include personalized phishing attacks, Cloud and SaaS attacks, as well as trojans — are more than twice as likely to become high-priority incidents as the average threat. With so much money on the line, every retailer should expect to confront targeted attacks throughout the weekend.

Bypassing the defenses

From ransomware to data exfiltration, one can make an educated guess about the kinds of threats facing retailers this Black Friday. But the truth is that no one knows exactly what the next global cyber-attack will look like, particularly given the enormous incentive for criminals to create an entirely new attack strain — or even a new type of attack altogether. Several recent, state-sponsored exploits have proven that the financial and technical backing exists to produce malware sophisticated enough to deliver a serious blow to the U.S. economy.

Innovative attacks pose a fundamental problem for traditional security tools, which rely on knowledge of past incidents to stop future ones. By updating their predefined notions of what constitutes a cyber-threat when a breach occurs, the best of these tools stop previously known attacks, but they are nonetheless blind to unknown threats. Many retailers have deployed Darktrace’s AI cyber security because it doesn’t presume to know what tomorrow’s attack will look like; rather, Darktrace learns on the job to differentiate between normal and abnormal behavior. But while such adaptive security is the only approach that stands a chance in today’s fast-changing threat landscape, most retailers have yet to make the switch.

In this era of DNA forensics and near-ubiquitous surveillance, the criminal heist has not disappeared — it’s digitized. And while retail companies prepare themselves for the generic cyber-threats of the past, very few are in a position to counter a never-before-seen attack that, like a physical heist, has been planned for months to exploit their unique security blind spots. As we inch closer to zero hour, the industry must be willing to adapt its cyber defenses against an ever-evolving adversary, or it may end Black Friday firmly in the red.

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
Justin Fier
SVP, Red Team Operations

More in this series

No items found.

Blog

/

Email

/

May 1, 2026

How email-delivered prompt injection attacks can target enterprise AI – and why it matters

Default blog imageDefault blog image

What are email-delivered prompt injection attacks?

As organizations rapidly adopt AI assistants to improve productivity, a new class of cyber risk is emerging alongside them: email-delivered AI prompt injection. Unlike traditional attacks that target software vulnerabilities or rely on social engineering, this is the act of embedding malicious or manipulative instructions into content that an AI system will process as part of its normal workflow. Because modern AI tools are designed to ingest and reason over large volumes of data, including emails, documents, and chat histories, they can unintentionally treat hidden attacker-controlled text as legitimate input.  

At Darktrace, our analysis has shown an increase of 90% in the number of customer deployments showing signals associated with potential prompt injection attempts since we began monitoring for this type of activity in late 2025. While it is not always possible to definitively attribute each instance, internal scoring systems designed to identify characteristics consistent with prompt injection have recorded a growing number of high-confidence matches. The upward trend suggests that attackers are actively experimenting with these techniques.

Recent examples of prompt injection attacks

Two early examples of this evolving threat are HashJack and ShadowLeak, which illustrate prompt injection in practice.

HashJack is a novel prompt injection technique discovered in November 2025 that exploits AI-powered web browsers and agentic AI browser assistants. By hiding malicious instructions within the URL fragment (after the # symbol) of a legitimate, trusted website, attackers can trick AI web assistants into performing malicious actions – potentially inserting phishing links, fake contact details, or misleading guidance directly into what appears to be a trusted AI-generated output.

ShadowLeak is a prompt injection method to exfiltrate PII identified in September 2025. This was a flaw in ChatGPT (now patched by OpenAI) which worked via an agent connected to email. If attackers sent the target an email containing a hidden prompt, the agent was tricked into leaking sensitive information to the attacker with no user action or visible UI.

What’s the risk of email-delivered prompt injection attacks?

Enterprise AI assistants often have complete visibility across emails, documents, and internal platforms. This means an attacker does not need to compromise credentials or move laterally through an environment. If successful, they can influence the AI to retrieve relevant information seamlessly, without the labor of compromise and privilege escalation.

The first risk is data exfiltration. In a prompt injection scenario, malicious instructions may be embedded within an ordinary email. As in the ShadowLeak attack, when AI processes that content as part of a legitimate task, it may interpret the hidden text as an instruction. This could result in the AI disclosing sensitive data, summarizing confidential communications, or exposing internal context that would otherwise require significant effort to obtain.

The second risk is agentic workflow poisoning. As AI systems take on more active roles, prompt injection can influence how they behave over time. An attacker could embed instructions that persist across interactions, such as causing the AI to include malicious links in responses or redirect users to untrusted resources. In this way, the attacker inserts themselves into the workflow, effectively acting as a man-in-the-middle within the AI system.

Why can’t other solutions catch email-delivered prompt injection attacks?

AI prompt injection challenges many of the assumptions that traditional email security is built on. It does not fit the usual patterns of phishing, where the goal is to trick a user into clicking a link or opening an attachment.  

Most security solutions are designed to detect signals associated with user engagement: suspicious links, unusual attachments, or social engineering cues. Prompt injection avoids these indicators entirely, meaning there are fewer obvious red flags.

In this case, the intention is actually the opposite of user solicitation. The objective is simply for the email to be delivered and remain in the inbox, appearing benign and unremarkable. The malicious element is not something the recipient is expected to engage with, or even notice.

Detection is further complicated by the nature of the prompts themselves. Unlike known malware signatures or consistent phishing patterns, injected prompts can vary widely in structure and wording. This makes simple pattern-matching approaches, such as regex, unreliable. A broad rule set risks generating large numbers of false positives, while a narrow one is unlikely to capture the diversity of possible injections.

How does Darktrace catch these types of attacks?

The Darktrace approach to email security more generally is to look beyond individual indicators and assess context, which also applies here.  

For example, our prompt density score identifies clusters of prompt-like language within an email rather than just single occurrences. Instead of treating the presence of a phrase as a blocking signal, the focus is on whether there is an unusual concentration of these patterns in a way that suggests injection. Additional weighting can be applied where there are signs of obfuscation. For example, text that is hidden from the user – such as white font or font size zero – but still readable by AI systems can indicate an attempt to conceal malicious prompts.

This is combined with broader behavioral signals. The same communication context used to detect other threats remains relevant, such as whether the content is unusual for the recipient or deviates from normal patterns.

Ask your email provider about email-delivered AI prompt injection

Prompt injection targets not just employees, but the AI systems they rely on, so security approaches need to account for both.

Though there are clear indications of emerging activity, it remains to be seen how popular prompt injection will be with attackers going forward. Still, considering the potential impact of this attack type, it’s worth checking if this risk has been considered by your email security provider.

Questions to ask your email security provider

  • What safeguards are in place to prevent emails from influencing AI‑driven workflows over time?
  • How do you assess email content that’s benign for a human reader, but may carry hidden instructions intended for AI systems?
  • If an email contains no links, no attachments, and no social engineering cues, what signals would your platform use to identify malicious intent?

Visit the Darktrace / EMAIL product hub to discover how we detect and respond to advanced communication threats.  

Learn more about securing AI in your enterprise.

Continue reading
About the author
Kiri Addison
Senior Director of Product

Blog

/

AI

/

April 30, 2026

Mythos vs Ethos: Defending in an Era of AI‑Accelerated Vulnerability Discovery

mythos vulnerability discoveryDefault blog imageDefault blog image

Anthropic’s Mythos and what it means for security teams

Recent attention on systems such as Anthropic Mythos highlights a notable problem for defenders. Namely that disclosure’s role in coordinating defensive action is eroding.

As AI systems gain stronger reasoning and coding capability, their usefulness in analyzing complex software environments and identifying weaknesses naturally increases. What has changed is not attacker motivation, but the conditions under which defenders learn about and organize around risk. Vulnerability discovery and exploitation increasingly unfold in ways that turn disclosure into a retrospective signal rather than a reliable starting point for defense.

Faster discovery was inevitable and is already visible

The acceleration of vulnerability discovery was already observable across the ecosystem. Publicly disclosed vulnerabilities (CVEs) have grown at double-digit rates for the past two years, including a 32% increase in 2024 according to NIST, driven in part by AI even prior to Anthropic’s Mythos model. Most notably XBOW topped the HackerOne US bug bounty leaderboard, marking the first time an autonomous penetration tester had done so.  

The technical frontier for AI capabilities has been described elsewhere as jagged, and the implication is that Mythos is exceptional but not unique in this capability. While Mythos appears to make significant progress in complex vulnerability analysis, many other models are already able to find and exploit weaknesses to varying degrees.  

What matters here is not which model performs best, but the fact that vulnerability discovery is no longer a scarce or tightly bounded capability.

The consequence of this shift is not simply earlier discovery. It is a change in the defender-attacker race condition. Disclosure once acted as a rough synchronization point. While attackers sometimes had earlier knowledge, disclosure generally marked the moment when risk became visible and defensive action could be broadly coordinated. Increasingly, that coordination will no longer exist. Exploitation may be underway well before a CVE is published, if it is published at all.

Why patch velocity alone is not the answer

The instinctive response to this shift is to focus on patching faster, but treating patch velocity as the primary solution misunderstands the problem. Most organizations are already constrained in how quickly they can remediate vulnerabilities. Asset sprawl, operational risk, testing requirements, uptime commitments, and unclear ownership all limit response speed, even when vulnerabilities are well understood.

If discovery and exploitation now routinely precede disclosure, then patching cannot be the first line of defense. It becomes one necessary control applied within a timeline that has already shifted. This does not imply that organizations should patch less. It means that patching cannot serve as the organizing principle for defense.

Defense needs a more stable anchor

If disclosure no longer defines when defense begins, then defense needs a reference point that does not depend on knowing the vulnerability in advance.  

Every digital environment has a behavioral character. Systems authenticate, communicate, execute processes, and access resources in relatively consistent ways over time. These patterns are not static rules or signatures. They are learned behaviors that reflect how an organization operates.

When exploitation occurs, even via previously unknown vulnerabilities, those behavioral patterns change.

Attackers may use novel techniques, but they still need to gain access, create processes, move laterally, and will ultimately interact with systems in ways that diverge from what is expected. That deviation is observable regardless of whether the underlying weakness has been formally named.

In an environment where disclosure can no longer be relied on for timing or coordination, behavioral understanding is no longer an optional enhancement; it becomes the only consistently available defensive signal.

Detecting risk before disclosure

Darktrace’s threat research has consistently shown that malicious activity often becomes visible before public disclosure.

In multiple cases, including exploitation of Ivanti, SAP NetWeaver, and Trimble Cityworks, Darktrace detected anomalous behavior days or weeks ahead of CVE publication. These detections did not rely on signatures, threat intelligence feeds, or awareness of the vulnerability itself. They emerged because systems began behaving in ways that did not align with their established patterns.

This reflects a defensive approach grounded in ‘Ethos’, in contrast to the unbounded exploration represented by ‘Mythos’. Here, Mythos describes continuous vulnerability discovery at speed and scale. Ethos reflects an understanding of what is normal and expected within a specific environment, grounded in observed behavior.

Revisiting assume breach

These conditions reinforce a principle long embedded in Zero Trust thinking: assume breach.

If exploitation can occur before disclosure, patching vulnerabilities can no longer act as the organizing principle for defense. Instead, effective defense must focus on monitoring for misuse and constraining attacker activity once access is achieved. Behavioral monitoring allows organizations to identify early‑stage compromise and respond while uncertainty remains, rather than waiting for formal verification.

AI plays a critical role here, not by predicting every exploit, but by continuously learning what normal looks like within a specific environment and identifying meaningful deviation at machine speed. Identifying that deviation enables defenders to respond by constraining activity back towards normal patterns of behavior.

Not an arms race, but an asymmetry

AI is often framed as fueling an arms race between attackers and defenders. In practice, the more important dynamic is asymmetry.

Attackers operate broadly, scanning many environments for opportunities. Defenders operate deeply within their own systems, and it’s this business context which is so significant. Behavioral understanding gives defenders a durable advantage. Attackers may automate discovery, but they cannot easily reproduce what belonging looks like inside a particular organization.

A changed defensive model

AI‑accelerated vulnerability discovery does not mean defenders have lost. It does mean that disclosure‑driven, patch‑centric models no longer provide a sufficient foundation for resilience.

As vulnerability volumes grow and exploitation timelines compress, effective defense increasingly depends on continuous behavioral understanding, detection that does not rely on prior disclosure, and rapid containment to limit impact. In this model, CVEs confirm risk rather than define when defense begins.

The industry has already seen this approach work in practice. As AI continues to reshape both offense and defense, behavioral detection will move from being complementary to being essential.

Continue reading
About the author
Andrew Hollister
Principal Solutions Engineer, Cyber Technician
Your data. Our AI.
Elevate your network security with Darktrace AI