Resources
Product Blog
Technical deep dives, architecture notes, and engineering write-ups from the SecureAuth product team.
AI Agents Security
11 posts
Agent authorization: let AI assistants do admin tasks for you
If an assistant uses a shared admin account and the log records only that account, you can't tell who asked. Calls to the Agent Authority management tools run with the roles of the person the assistant belongs to, and the audit log records each call with both the assistant and that person.
AI agent least privilege as MCP servers and tools change
Teams add MCP servers, and servers add tools. AI agent access control written as lists of tool names falls behind. Agent Authority denies tools by default and allows them by tag, so a new tool is covered by its tags.
AI agent access control: deny rules and usage limits
Some AI agent actions should never happen. Others are acceptable one at a time and risky in volume. Agent Authority handles the first with deny rules and the second with usage limits on tool calls.
Mask sensitive data in the answers your AI agents get back
Access rules decide on a call before the business system answers. An account number in a case note can still come back, and Agent Authority's data redaction and masking can hide it before the agent sees it, when it matches a pattern you configure.
Beyond the MCP gateway: safe AI tool access for every employee
Enterprises are standing up MCP gateways to connect AI to their internal tools. A gateway is a sensible first step. But if that is where you stop, you get neither the productivity nor the safety you were after.
How MCP Authority gives employees safe access to AI tools
MCP Authority is five layers working as one system. This walkthrough shows each of them in Agent Authority, and follows a single tool request from the catalog to an approved, policy-checked, audited call.
Agent API security: FAPI 2.0 is the floor
The software calling your APIs used to be a browser or a mobile app. More and more, it is an AI agent, acting for a person and making calls on its own, without anyone watching each step. That means route checks and bearer tokens are no longer enough on their own.
Your IdP authorized the connection, not the action
First in a three-part series. EMA and ID-JAG are a genuinely good open standard for governing the door, and they were never built to decide whether a specific action should run. Here is exactly where connection governance stops.
The Agent Authority pattern
Part 3 of 3. If SCIM cannot be the kill switch, something on your side of every vendor boundary must be. Agent Authority: token custody, per-call ticket exchange, and revocation measured in milliseconds.
Your agent does not log in. SCIM does not know that.
Part 2 of 3. SCIM is a wire protocol, not a guarantee, and your blast radius is the worst vendor in your stack. Why SCIM-based revocation cannot be the kill switch you need for agent tokens.
The new JML gap: agents outlive their humans
Part 1 of 3. Deprovisioning revokes the human in Okta, but the agent runtime on a personal laptop keeps refreshing cached tokens long after Monday. The new shape of the Joiner-Mover-Leaver problem.
AI Agent Adoption
1 post
MCP Gateway
2 posts
Beyond the MCP gateway: safe AI tool access for every employee
Enterprises are standing up MCP gateways to connect AI to their internal tools. A gateway is a sensible first step. But if that is where you stop, you get neither the productivity nor the safety you were after.
How MCP Authority gives employees safe access to AI tools
MCP Authority is five layers working as one system. This walkthrough shows each of them in Agent Authority, and follows a single tool request from the catalog to an approved, policy-checked, audited call.
Architecture
2 posts
MFA for critical infrastructure networks that can't reach the internet
Power grids, pipelines, water treatment and defense systems still need strong MFA when their networks can't reach the cloud. Here's how SecureAuth secures privileged access to critical systems, entirely inside the perimeter.
FAPI 2.0 certified: putting consumers in control of their data
When your bank lets a budgeting app see your transactions, you should be the one who says yes, decides how long that access lasts, and can take it back whenever you want. Open banking made that the expectation. FAPI, short for Financial-grade API, is how it actually works.
Passwordless & MFA
4 posts
Beyond Frictionless: Risk-Based Passwordless Authentication
Risk-based authentication already let trusted users skip MFA. With SecureAuth release 26.4, the same risk score can take the password away too, and adaptive MFA chooses a stronger method only when a sign-in looks risky.
Moving your employees to passkeys? Keep real-time risk scoring
The FIDO Passkey Index 2026 puts passkeys at 36% of sign-ins at the best-resourced deployments in the world. The other 64% of sign-ins, the enrollment that creates a passkey and the session after the login still need a risk score.
Lower SMS costs with risk-based MFA
Most MFA policies send a code on every login, and each SMS code has a price. We modeled risk-based step-up on three months of real logins from a large US health insurer, and invisible MFA verified 85% of returning logins without a challenge.
Control YubiKey registration at the device level with Enterprise Attestation
FIDO2 is phishing-resistant, but at registration every YubiKey of the same model looks the same. Enterprise Attestation lets SecureAuth check each key's serial number against your approved list and block keys that aren't on it.
Product Updates
8 posts
Agent authorization: let AI assistants do admin tasks for you
If an assistant uses a shared admin account and the log records only that account, you can't tell who asked. Calls to the Agent Authority management tools run with the roles of the person the assistant belongs to, and the audit log records each call with both the assistant and that person.
AI agent least privilege as MCP servers and tools change
Teams add MCP servers, and servers add tools. AI agent access control written as lists of tool names falls behind. Agent Authority denies tools by default and allows them by tag, so a new tool is covered by its tags.
AI agent access control: deny rules and usage limits
Some AI agent actions should never happen. Others are acceptable one at a time and risky in volume. Agent Authority handles the first with deny rules and the second with usage limits on tool calls.
Product Radar: September 2026
Everything we shipped in September, in one place. Shadow agent and NHI discovery, identity risk for Agent Authority, use your AI assistant, password breach detection and adaptive MFA for the workforce, a RADIUS Server security update, Presence Authority early access and passkey setup at activation.
Product Radar: August 2026
Everything we shipped in August, in one place. Rate limits and per-service grants for AI agents, delegated application assignment and embeddable sign-in for B2B and CIAM, risk-based prompts for the workforce, and more.
How we let agents reach real enterprise systems without over-permissioning them
A product update on recent weeks. Microsoft 365 Graph reached GA with SharePoint, OneDrive, and Teams access, eight connectors joined the catalog, and cost is now attributed by team. The theme is reach that stays governed.
Product Radar
Everything we shipped, in one place. Agent Authority for securing and governing AI agent access to enterprise systems, plus lower OTP costs on your own carrier, OTP fraud blocked by default, passkeys that survive a domain change, scoped Dashboard access and native OpenLDAP.
Agent Radar: MCP Gateway
Once an agent can act on your systems, who makes sure it does only what it should? Agent Authority exists to enforce that boundary: a runtime control plane and governance layer for deploying agents securely across an organization.