Skip to main content
Gartner® identifies agentic as a top-priority emerging technology on the Global Attack Surface Grid.
SecureAuthSecureAuth
← Product Blog

Product Updates

Product Radar: August 2026

Mahnaz Majeed11 min read
On this page

The agent is a new identity type. We are building it as one rather than bolting it onto the identity types we already support.

Our Agent Authority releases this month build agent identity on open standards and API-first foundations, on top of the authentication and authorization we already run at scale:

  • RFC 8693 token exchange for cross-application hand-off, so an agent carries a verifiable identity from one application to the next.
  • Client identity published as a metadata document, rather than registered in advance, so agents can come online without a manual registration step.
  • Rate limits and data redaction enforced at the gateway, so a misbehaving agent cannot drain an API or return data the requester should not see.
  • Policy evaluated before access, so what an agent may reach is decided per request rather than inherited from a standing grant.

Featured this month

Presence Authority

Stop IP theft and data exfiltration through the employee’s desktop screen.

Faster accurate detection with far fewer false positives

Human detection is now stronger. The engine tracks moving body parts rather than the face alone, which widens the range of activity it can recognize and reduces false alarms.

It now detects:

  • Unauthorized screen capture

    Capture activity at the workstation is flagged while the session is live.

  • Phone cameras pointed at the screen

    The motion of raising a device into frame registers even when the operator stays seated and facing forward.

  • Shoulder surfing

    A second person moving into view behind the operator raises an alert.

A laptop enclosed by a shield, with five numbered exfiltration paths out of it, screen capture, screen share, phone camera, bystander and idle session, each marked closed.

Agent Authority

For access control, security, governance, and audit over AI agents and non-human identities

  • More governed connections: expand what agents can reach under the same guardrails. This release widens coverage on both sides of the gateway:

    • New supported AI agents. Gemini CLI and Antigravity.
    • New systems an agent can reach through us. HubSpot, Ramp, Lucid, and G2.
    • Extended Microsoft coverage. Now reaches Teams chats and channels.

    Each addition runs under the policy and audit trail already in place, so the list of connected systems grows without a separate access control model per connection.

    A policy graph linking teams on the left through the AI agents they use, such as Claude, Cursor, Codex and Gemini CLI, to the allow and deny policies that govern them, and on to tagged groups of systems.
    One graph to show you every team, every agent and every policy that governs them.

  • Bring your own AI provider: keep data-protection checks on a model you control. The AI judge can now run on a provider and model you choose:

    • Register your own AI provider. Judgments run through your provider account rather than a shared default.
    • Select the model. You decide which model performs Data Protection judgments.
    • Fails closed when unconfigured. If no provider is set, the check refuses the request instead of allowing the data through unchecked.
    A data redaction rule that redacts responses for named assistants on Salesforce and DocuSign, combining an exact pattern match filter with an AI judge prompted to flag full names, Social Security numbers, home addresses and dates of birth.
    Your chosen AI model can also perform Data Protection redaction.

  • Webhooks: stream agent audit events into your own tools in real time. Gateway audit events can now reach your own systems as they happen:

    • Send to the tools you already run. Slack, Teams, PagerDuty, CrowdStrike, Datadog, Splunk, or any HTTPS endpoint.
    • Filter before delivery. Choose what goes where by event type or condition, so each destination receives only the events it acts on.
    • Set up webhooks yourself. Creating a webhook is self-serve, so there is no need to ask us for support and no waiting for a feature to be turned on.
    A destination picker for audit event webhooks offering Slack, Microsoft Teams, PagerDuty, CrowdStrike, Datadog, Splunk HEC, and a custom HTTPS endpoint.
    Gateway audit events can be routed to the tools a security team already runs.

  • Rate limits on tool calls: contain runaway or abusive AI agent activity before it costs you. Rate limits now apply to agent tool calls:

    • Set one limit or many. Apply caps across the board, or scope per user, group, system, or tool.
    • Preview before enforcing. Log-only mode records what would have been refused, so you can size the limit before it takes effect.
    • Runaway agents stop at the gateway. A looping agent is refused before the call reaches the downstream system, and each refusal lands in the audit trail.
    A rule editor throttling tool calls by Claude and ChatGPT on Salesforce soqlQuery, budgeted to ten calls per minute counted per user and agent instance, refusing the call when the budget runs out.
    Throttling an agent to ten Salesforce queries a minute, counted per user and agent instance.

  • Tools grouped by the service they come from: grant an agent only the services it needs, not whole connectors. Multi-service connectors are now split by service:

    • Connect one service at a time. Google Workspace and Microsoft 365 appear as their individual services when you set up the connection.
    • Write policy per service. A rule names the service it applies to, not the connector as a whole.
    • A grant stops at the service named. Approving access to Gmail does not also approve Drive.

  • AI spend by team, model, and person: see and control AI cost by team and person, not just org-wide. AI spend now breaks down by team and by individual:

    • Use the groups you already have. Spend maps to your existing groups, so there is no separate reporting structure to set up.
    • Compare licensed and active users. See how many seats are paid for compared to active users.
    • GitHub Copilot is included. Copilot now appears in the cost view alongside the other assistants.
    A cost dashboard showing adoption versus cost per active user, seat usage by team, and model mix broken down by team with a dollar total for each.
    Spend broken out by team, by model and by whether licensed seats are being used.

  • Connect an agent to one system, not everything: keep agents focused and within their tool limits. The connect dialog can now be narrowed to a single system:

    • Narrow to one system. Select a single system and the agent loads only that system’s tools.
    • Shorter tool lists. A person sees only the tools for the system they are working in, rather than the full set from a broad connection.

B2B Authority and Customer Authority

For partner and B2B access, delegated administration, consumer sign-in, and advanced API Security

  • Delegated application assignment for B2B organizations and groups: onboard and offboard whole partner organizations in one step. Access can now be granted at the organization level:

    • Grant to a whole organization or group. Assign an application once rather than to each user in turn.
    • Revoke the same way. Removing access at the organization level removes it for everyone it covered.
    • The partner administers their own users. An admin at the partner organization decides which of their staff receive access.
    An application's Assignments tab listing the organizations that can reach it, each with a mode of Direct, Inherit or Delegate.
    Grant the application to an organization and let that organization's own admin manage it from there.

  • Embeddable sign-in and self-service components: sign-in and account management can now run inside your own pages:

    • Embed the screens you need. Sign-in, profile, and account security are available as components you place in your own pages.
    • They match your design. The components are styled to your site rather than to a separate hosted page.
    • No redirect. Customers stay on your site through sign-in and through account changes.

  • Bot protection on the activation link: activation links now survive email security scanning:

    • A challenge protects the activation code. The code is accepted only after the user accepts the challenge.
    • Scanners no longer consume the link. A corporate link scanner following the link does not complete the challenge, so the code stays valid for the user.
    • Users activate on the first click. New users no longer need to request a replacement link or submit a support ticket for an expired activation link.

  • Cross-application hand-off (ID-JAG): users can move between partner apps without signing in again:

    • Accept a hand-off token from an external provider. The token is short-lived and names both the user and the one app being entered.
    • Exchange it for access. The user enters that app with no second sign-in.
    • Invalid hand-offs are refused. A hand-off that names no user, or that names a different app, is refused.

    You can put this to work with Agent Authority today rather than reading it as a specification: see how the hand-off runs end to end.

  • Client ID Metadata Documents: onboard applications and AI assistants at scale without pre-registration. An application is identified by an HTTPS address that publishes its own details, instead of being registered in advance. When you open an API to many AI assistants, each one publishes for itself and its record is created the first time it is approved.

Workforce Authority

For employee and contractor sign-in across your workforce

  • Intelligent assurance-based authentication flows (coming soon): the authentication prompt is now decided by risk:

    • The decision is made per sign-in risk level. Risk shapes what the user is asked for, rather than one policy applied to everyone.
    • Low-risk sign-ins move through faster. Less friction for sessions that score low.
    • The strongest method is held for higher risk. Users are asked for it when the session warrants it.

  • Administrative control and visibility (coming soon): admins can now clear blocks themselves:

    • Release a locked-out user. Restore access without waiting out the lockout timer.
    • Clear a wrongly blocked IP. Remove the block directly rather than waiting for it to expire.
    • Every action is recorded. The audit record shows who did it and why.

  • Improved user access and login experience (coming soon): sign-in asks less time of users, and fewer of them call the help desk:

    • Voice OTP for help desk verification. The help desk can verify a caller by voice OTP, with no phone app required.
    • Self-service in the user’s own language. Screens appear in the language the user has set.
    • No repeat prompt after federated MFA. A user who already satisfied MFA at their own provider is not asked again.

Assurance Authority

For advanced risk intelligence across authentication

  • Enhanced risk intelligence dashboard: risk reporting now also shows how authentications fall against your threshold:

    • See the split above and below the threshold. The report shows how many authentications landed on each side of your current setting.
    • Tune against real traffic. Adjust the threshold using your own volumes rather than an estimate.

Presence Authority

For continuous, camera-based session assurance: the right person, a safe environment, no visual data loss

  • Pose tracking: fewer false alarms when users glance away. A new upper-body model keeps verification continuous when the face is briefly out of view:

    • Presence is confirmed from upper-body position. When the face leaves the frame for a moment, the model continues to verify the operator.
    • Fewer interruptions at the desk. A brief turn away no longer raises a false alarm.
    • Admins keep control of the behavior. Run the model in shadow mode before enforcing it and set a limit on how long pose is tracked.

Questions or want to know more?

Reach the product team directly at product@secureauth.com. We’d love to hear what you’d like to see next or if you want to chat with us.