A prompt can reveal what someone is asking an AI system to do, but it cannot establish on its own whether that activity is safe or suspicious. Security teams need to interpret prompt language alongside the identity behind it, normal behavior, connected systems and data, relevant collaboration, and what happened next.
Why prompt language matters—and why it is not enough
Employees now use chatbots, copilots, coding assistants, and autonomous agents for work. Their prompts can offer a useful window into how those systems are being used. As Nabil Zoldjalali, Darktrace’s VP, Field CISO, writes, “Prompt analysis will undoubtedly become more common, as prompts are one of the clearest windows into how people and agents are using AI systems.”
As an Amazon Associate I earn from qualifying purchases.
But a prompt records a request, not its full circumstances. Similar wording might be routine for one employee, unexpected for another, or concerning if it comes from a compromised account. The systems and data available to the requester—and the actions taken after the interaction—can change how the same language should be understood.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →In his June 24, 2026 article, Zoldjalali argues that prompt inspection should be interpreted within broader enterprise security context. That is a vendor-authored strategic position, not an independently measured finding about detection accuracy or product performance.
#1 Best Overall
How context changes the meaning of activity
A rise in AI use may have an ordinary explanation
Zoldjalali offers a hypothetical employee facing a deadline. Their AI use, document work, and interactions with enterprise systems all increase. Seen in isolation, the change could resemble insider risk or use of an unmanaged AI workflow. If the employee is completing a legitimate, time-sensitive assignment from a senior leader, and their collaboration patterns fit the work, the activity may be expected.
This is an illustrative scenario, not a documented incident or measured case study. Its point is that a behavioral change is a reason to investigate, not a verdict.
Ordinary wording can still merit scrutiny
The reverse is also possible: routine-sounding language may deserve attention if the person or agent behind it is compromised, unfamiliar, outside its usual responsibilities, or using a shadow AI workflow. Prompt wording alone may not reveal those circumstances.
What security teams should examine alongside prompts
Zoldjalali’s proposed approach is to connect language signals to the surrounding activity. When reviewing an AI interaction, teams can ask:
Rank #3
- Who or what issued it? Identify whether the prompt came from a person, an agent, or another workflow.
- What is normal for that identity? Consider the requester’s usual behavior across the enterprise, not only their prompt history.
- What could the interaction reach? Identify connected systems, data, and workflows relevant to the request.
- What relationships or communications explain the timing? Collaboration may clarify whether an unusual burst of work fits a real assignment.
- What happened afterward? Check whether downstream actions aligned with the expected business task.
This is context suggested by the article, not a standardized or independently validated checklist. Organizations would need to apply it within their own security policies, access controls, and operational needs.
Prompt inspection and contextual analysis are different approaches
| Approach | What it can help surface | What it cannot establish by itself |
|---|---|---|
| Prompt-only inspection | The language and apparent request in an AI interaction. | Whether the requester is authorized, whether the behavior is normal for them, what systems or data were involved, or whether later actions fit the task. |
| Contextual correlation | How identity, behavior, connected resources, collaboration, and subsequent actions relate to the interaction. | It does not make intent certain; context still needs interpretation and investigation. |
This comparison describes the approaches in Zoldjalali’s commentary; it is not a measured ranking of security products. His article also argues that perimeter, identity, and data-security perspectives each contribute, but none alone explains the full operational context of AI use.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the source does—and does not—demonstrate
The article makes a qualitative case for interpreting prompts in context. It does not report a controlled evaluation, detection or false-positive rates, a cost comparison, or an independent comparison of products. Its recommendation should therefore be read as the author’s strategic view, not as proof that a particular method or tool detects misuse more accurately.
The article’s caption describes a reconstruction of user and agent interactions with risky prompts highlighted and categorized, and names a Darktrace offering. That is a vendor product description; it does not establish independent performance results.
Best Value
A practical principle for enterprise AI security
Use prompt language as one signal among several. An unusual request or change in usage can help direct attention, but understanding risk calls for connecting that signal to the identity, access, business relationships, and actions around it. As Zoldjalali puts it, “The future of prompt analysis is not just about understanding language. It is about understanding language in context.”
Read the original Darktrace article by Nabil Zoldjalali, published June 24, 2026.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
