Enkrypt AI, by Anaconda, integrates with the OpenAI Compliance API. Our guardrails scan your ChatGPT Enterprise workspace activity on a recurring schedule and give you a regularly updated view of its compliance posture.
Here’s the problem we kept running into with customers.
ChatGPT Enterprise gives admins solid controls and auditable records. But the workspace analytics are built around adoption, not risk. #
Open the admin console and you get a clear picture of usage: seats assigned, message volume climbing, which custom GPTs teams built and which ones caught on. Those metrics all answer one question: is this being used?
Security teams are actually asking a different one: is this being used safely?
Usage inside a ChatGPT Enterprise workspace isn’t one thing anymore. Someone pastes a client roster into a chat to reformat it. Someone uploads a contract PDF and asks for a summary, with no idea what instructions are buried in the document. A developer runs a Codex session against a private repo. A workspace agent fires off a connector call into Drive or SharePoint.
Some of that is completely fine. Some of it breaks your AI usage policy. Some of it is someone testing how far they can push the model. In adoption metrics it all looks identical: a message count going up.
The Compliance API does give you the underlying record, and the coverage is broad: conversation messages, Codex activity, agent runs, connector calls. But it’s raw material, not an answer. You get thousands of immutable, gzipped JSONL event files, and Enkrypt AI automates review against the customer’s policies and helps surface findings for investigation.
Most security teams don’t have the time to do that. So potential risk sits in the logs, unreviewed. It tends to come up at the worst moment: an audit, an incident, or a regulator asking for evidence nobody collected.
Retrospective monitoring, on your schedule #
Enkrypt AI reads your workspace activity from the OpenAI Compliance API and runs it through the same guardrails our customers use in production. It flags potential issues after the fact and doesn’t block anything, and your admins stay in control of remediation.
We scan supported activity from the sources you enable, including prompts, responses, tool calls, and uploaded files with extractable text, and roll the results up into one view of where your workspace stands.
To get started, you need an authorized Enkrypt AI account, a Compliance API key with the right scopes, and your workspace ID. Then pick a time range, sources, and the guardrails, and run it. Nothing changes for your employees: no plugins, no browser extensions, nothing added to anyone’s workflow. The monitoring runs on the compliance side, where it belongs.
What we watch for #
Prompt injection. Attempts to override instructions, jailbreak the model, or manipulate it into unsafe behavior, including injections buried inside uploaded documents.
Policy violations. Your policies, not ours. Define your organization’s AI usage rules on the Enkrypt AI platform and we flag the specific rule that may have been violated, whether that’s data exfiltration, restricted topics, or anything else you care about.
Sensitive-data exposure. Personal and sensitive information moving through your workspace, whether an employee pastes it into a prompt or a model returns it in a response, flagged against your own data policies.
Toxicity. Harassing, abusive, or offensive language in prompts and in responses.
Not safe for work (NSFW) content. Explicit or inappropriate material anywhere in the interaction.
Coverage that goes where your teams go #
ChatGPT Enterprise has grown well past a single chat window, so monitoring it properly means following the work. We scan ChatGPT Enterprise conversations, Codex sessions, workspace agents, connectors, and the spreadsheet plugin. Turn them all on, or narrow the scan to what matters most.
We also go deeper than the message text. When someone uploads a file mid-conversation, we pull its contents out of the logs and scan those too. In one scan of a customer’s workspace, a user uploaded a PDF and asked the model to follow its instructions. The document carried a prompt injection. We extracted the text, ran it through our guardrails, and flagged it. Risk hidden inside file contents is exactly the kind of thing scheduled monitoring exists to catch.
Your compliance landscape, on one screen #
Every scan produces a single view of where your workspace stands.
Headline numbers first: injection attempts, policy violations, total flags, and your overall flag rate. Around them, the usage context that makes those numbers mean something, including conversation volume, active users, which models and services are in play, and how many uploaded files were scanned.
Then the detail that turns a number into a decision:
- Per-detector breakdown, so you can see at a glance whether your exposure is concentrated in injection attempts or policy drift
- Policy violations by rule, so you can see which rules drive the most flags instead of guessing
- Compliance framework mapping that ties findings to the relevant articles, controls, and categories in the European Union AI Act ,International Organization for Standardization (ISO) and International Electrotechnical Commission (IEC) 42001 ,ISO/IEC 24028 , MITRE Adversarial Threat Landscape for Artificial-Intelligence Systems (ATLAS),National Institute of Standards and Technology AI Risk Management Framework (NIST) AI RMF, and the Open Worldwide Application Security Project (OWASP)Top 10 for Large Language Models (LLM) and Agentic Applications , so your evidence is already in the language your auditors use
- Per-service detection, showing whether risk lives in chat, in Codex, or somewhere you weren’t looking
- Inbound versus outbound, separating what your people are sending in from what the model is sending back
- Top users by flags, so patterns of repeated behavior surface on their own
Every finding comes with its evidence #
Open any finding and read the flagged message in full, with the key surrounding context and the conversation ID attached for your records. That’s the difference between a metric and evidence. You’re not asking a team to trust a count. You’re showing them what happened.
Move fast, and know where you stand #
Governance should never be the reason a company slows down its AI adoption. It should be the reason it can speed up.
With Enkrypt AI monitoring your workspace through the OpenAI Compliance API, you stop guessing about your compliance posture and start building a real record of it. When an auditor asks questions, you have documentation to point to. And instead of waiting for an audit to surface a problem, scheduled scans flag potential policy violations and manipulation attempts in supported activity, with the context your team needs to investigate.
ChatGPT Enterprise gave your organization a powerful new way to work. We help you document that activity and investigate potential policy violations.
See it on your own workspace #
If your organization runs ChatGPT Enterprise, we’ll show you how to review supported activity in your workspace. You’ll need an authorized Enkrypt AI account, a scoped Compliance API key, and your workspace ID. Reach out to the Enkrypt AI team for a walkthrough with your own policies in place. Request a walkthrough.
FAQ #
What does Enkrypt AI do for ChatGPT Enterprise?
Enkrypt AI connects to the OpenAI Compliance API and scans your ChatGPT Enterprise workspace activity against your organization’s AI usage policies. It flags potential risks and gives you one view of your workspace’s compliance posture.
Does Enkrypt AI block or change what employees can do in ChatGPT Enterprise?
No. Monitoring is retrospective and runs on the compliance side. It flags potential issues after the fact and doesn’t block anything. Employees don’t need plugins or browser extensions, and their workflow stays the same.
What do I need to get started?
You need an authorized Enkrypt AI account, a Compliance API key with the right scopes, and your workspace ID. After that, you choose a time range, the sources to scan, and the guardrails to apply.
How often does Enkrypt AI scan my workspace?
Scans run on a recurring schedule, so your view of the workspace’s compliance posture stays up to date.
Which parts of ChatGPT Enterprise does Enkrypt AI scan?
It scans ChatGPT Enterprise conversations, Codex sessions, workspace agents, connectors, and the spreadsheet plugin. Within those, it covers prompts, responses, tool calls, and uploaded files with extractable text. You can scan everything or only the sources you choose.
What types of risk does Enkrypt AI detect?
It detects prompt injection, violations of your own policies, sensitive-data exposure, toxicity, and NSFW content.
Can I use my own AI usage policies?
Yes. You define your organization’s rules on the Enkrypt AI platform, and Enkrypt AI flags the specific rule that may have been violated.
How does this help with audits?
Findings are mapped to the EU AI Act, ISO/IEC 42001 and 24028, MITRE ATLAS, NIST AI RMF, and the OWASP Top 10 for LLM and Agentic Applications. Each finding also includes the flagged message, the surrounding context, and the conversation ID, so you have evidence to show auditors instead of just a count.
Can Enkrypt AI detect risks hidden inside uploaded files?
Yes. When someone uploads a file with extractable text, Enkrypt AI pulls the contents out of the logs and scans them. This includes prompt injections hidden inside documents.