Questions we get from security and AI teams
Short, plain answers. If yours isn't here, ask it on a call.
Fit and coverage
Microsoft and AWS now have agent registries. Why do we need Aegis?
A registry lists agents that someone registered or synced. Aegis finds agents and AI tools nobody registered, across your clouds, SaaS apps and employee laptops, and scores what each can reach. Its main job starts after that: a policy decision before risky actions, and a record of it.
We already use Wiz or another cloud-posture tool. Why add Aegis?
Keep it. Cloud-posture tools assess how your cloud is configured. In the agents you build, Aegis decides on each wrapped tool call before it runs (allow, mask, hold for approval, or block) and records the decision with the AIUC-1 requirements it maps to. Aegis connects read-only, so the two run side by side.
We have SSE and DLP. Does that cover employee AI use?
Keep those too. Aegis adds an inventory of the AI apps, accounts and extensions employees use, plus coding agents and the MCP servers running on laptops, with your block, warn and masking rules applied on the device. It also puts policy inside the agents you build, which network tools don't reach.
What do you cover today?
Discovery: AWS (Lambda, ECS, EC2, EKS, Bedrock Agents), Azure (Functions, AKS, Virtual Machines, AI Foundry and Azure OpenAI), Google Cloud (Cloud Run, Cloud Functions, GKE, Compute Engine, Vertex AI Agent Engine), Salesforce Agentforce, ServiceNow, Microsoft Copilot Studio and Workday, and code you link. Laptops and browsers: macOS and Windows, Chrome and Edge. Runtime control: the Python agents you build, including LangChain, LangGraph, LlamaIndex, CrewAI and MCP. Agents inside SaaS apps such as Agentforce or Copilot Studio get discovery and risk scoring, not runtime control. Ask us about other platforms.
Do you cover the AI tools our employees use?
Yes. A managed browser extension and endpoint agent find the AI apps, accounts and extensions employees use, plus coding agents and local MCP servers. Your policy decides whether an unapproved tool is blocked or the employee sees a warning, and customer, health, financial and secret data pasted or uploaded into AI tools is detected or masked on the device.
How do you avoid false positives?
Every point of confidence traces to stored evidence. A key name or a GPU alone makes an agent "possible," never "confirmed." In our benchmark of 40 workloads, controls produced zero false positives. Your assessment readout shows the evidence behind each finding.
Security and data
What access does discovery need? Do you install anything?
Nothing is installed in your cloud or SaaS apps. For AWS you create one read-only IAM role; Aegis uses short-lived credentials with an ExternalId unique to you and never holds long-term keys. For Salesforce you create an External Client App. Azure, Google Cloud and the other SaaS apps also connect read-only. We never run your agents or write to your systems. Laptops are the exception: they run a managed browser extension and endpoint agent.
What leaves our environment, and where is it stored?
From clouds and SaaS apps: names and settings, not values; values are dropped on arrival, and prompt text and business data are not read. From laptops: metadata only, because detection and masking run on the device. From your agents: decisions and traces, with personal data and secrets redacted in your process first; you choose the capture mode (none, metadata only, or full). In our cloud, your data sits in its own database in Google Cloud US regions, and you set how long it is kept.
Does Aegis need write access to fix permissions?
Not in your cloud or SaaS apps. IAM fixes go out as a pull request to your infrastructure code, for your team to review and merge. Connector scope changes go to their owner as a request. Only Aegis policy changes take effect directly.
What happens if Aegis is down?
The call does not run. Unknown tool, evaluation error, or unreachable decision service all mean block. If a specific tool must keep working, you can set open mode for that tool, and every call made that way is recorded as a gap.
Can it run inside our network?
Yes. Aegis runs as SaaS on Google Cloud with a separate database per customer, or as a single-tenant appliance inside your network. The appliance needs no outbound connection; signed updates and counts-only usage reporting are optional.
Are you certified? Is AIUC-1 a certification?
Aegis is SOC 2 Type II; the report is available under NDA. Separately, Aegis maps runtime decisions to AIUC-1, a security standard for AI agents, and supports your work against NIST AI RMF and the OWASP Top 10 lists for LLM and agentic applications. That evidence supports an audit; it is not a certification.
For engineers
Do we have to change code?
Not for discovery. For runtime control, you install the Python SDK (pip install aegislib), create a client, and wrap each tool with a decorator. Policy changes reach your agents without a redeploy. See the developer details.
Will the SDK slow down our agents?
Measured overhead on the SDK path is 0.22 ms at p99 per policy check, with the network mocked. The round trip to the decision service depends on where it runs, in Aegis SaaS or in the appliance inside your network, and we measure it with you in a technical walkthrough.
How do approvals reach a person?
The call waits while approvers review the exact parameters in the Aegis console. You can require more than one approver; one rejection denies. Delivery to Slack, Teams and email is next.
Getting started
How long until we see findings?
Within an hour of connecting an AWS account or Salesforce org. Cloud discovery installs nothing: you create the read-only role or the External Client App, and the scan starts. Runtime control takes longer, because your team adds the SDK to each agent's code.
What does it cost?
We size pricing to your estate and share it on the first call. The agent assessment is free.