Snapshot 127368
Normalized text
Scripts and page chrome removed; this is what change detection compares.
NewPermit Nexus PDP benchmarks: 5.6 ms P99 on RBAC checks, measured server-sideNexus PDP: 5.6 ms P99 RBAC checksSee the benchmarks Least privilege that survives contact with the codebase Security writes the access policy. Engineering implements it in fifty places, slightly differently each time. Permit puts the policy in one layer that every service and agent has to ask, and gives you the log of every answer. Book a demoStart free denyagent:ops-assistanttools/callrepo:payments-service “At Epsagon (acquired by Cisco) we are no strangers to the complexity of microservices. Access control demands of microservices are never-ending, so they require a modern stack that can quickly adapt to the most demanding tech and security needs.” Ran RibenzaftEpsagon CTO, Cisco Why access policy drifts from reality Policy lives in documents, enforcement lives in code The approved access model and what the application actually does diverge with every release, and nobody can diff them. Standing access accumulates Service accounts, API keys, and broad roles pile up, because narrowing them always requires a code change nobody has time for. AI agents inherit too much Agents act through human credentials or long-lived tokens, so a single prompt injection gets whatever those credentials can reach. Incidents without answers After an incident, the first question is who could access what, and why. Application logs rarely say. How Permit gives security control back Authorization becomes a control you own and can inspect, instead of an implementation detail spread through the code. One policy, enforced everywhere Services, gateways, and agents call the same decision points, so the policy you review is the policy that runs. Default deny, narrowly granted RBAC, ABAC, and ReBAC grant access per resource and per condition, instead of per system. Zero standing permissions for AI agents Through the MCP Gateway, agents never hold real credentials. They act on a person’s live permissions, and revoking the person revokes the agent on its next call. A human in the loop where it matters Approval flows require sign-off before elevated access or high-risk actions, including actions requested by agents. Access control for your access control Policy guards decide who can change which policies, and with GitOps enabled, policy changes land in your Git repository with their history. Decision logs for detection and response Every decision is recorded with its reason and forwardable to your SIEM, so an investigation starts with facts. An agent action, stopped at the tool call An agent tries a destructive action beyond what its user consented to. Illustrative data. Example decisiondeny subject agent:ops-assistant for user:lee@example.com action tools/call delete_repository resource repo:payments-service reason consent covers read and open_pull_request only; delete needs lee’s approval decided by Permit MCP Gateway A decision point at every layer: people, services, and agents are checked at the gateway, the service, and the data, with every answer on record. One agent action, four enforcement points AI agentacting for a person MCP gatewayMay this agent call this tool, with these arguments?policy check API gatewayMay this delegated token reach this endpoint?policy check ApplicationMay this user approve this record in this tenant?policy check DatabaseMay this query read this column?policy check Each layer asks its own question of the same policy and the same identity. A prompt injection that slips past one check still meets the next. In their words “At Granulate we optimize our customers’ most critical systems; as a result, getting access control right is of the highest importance. Full stack permissions as a service allows our developers to focus on their core product. I was extremely impressed both by Permit.io’s technology and its dedication to customer service.” Tal SaiagFounder & CTO, Granulate Read the docs, run the code Documentation Policy decision pointsDeploying PDPs inside your own network. Policy guardsControl who can change which policies. Audit and decision logsWhat is recorded for every decision, and how to forward it. MCP GatewayAuthorization, consent, and audit for AI agent tool calls. Example projects on GitHub Secure AI agents with PydanticAIFine-grained authorization around an AI agent’s prompts, data, and actions. Four-perimeter AI access controlThe AI access control framework, demonstrated with Langflow. OPALThe open-source layer that keeps policy engines in sync. Questions security teams ask How does Permit reduce blast radius? By replacing broad, standing access with default-deny policies evaluated on every request. A compromised account, token, or agent can reach only what policy allows for that specific request. How are AI agents authorized? The MCP Gateway gives each agent an identity bound to the person it acts for, their consent, and its declared intent. Every tool call is checked against policy, and upstream credentials stay vaulted in the gateway. Who can change authorization policy? Policy guards restrict which people and teams can change which policies. With GitOps enabled, policy changes are committed to your Git repository, so they can be reviewed like any other code. Can we send decision logs to our SIEM? Yes. Decision logs include the user, action, resource, outcome, and reason, and can be forwarded to your SIEM or SOC tooling. Is Permit secure enough to sit in the access path? Permit holds a SOC 2 Type II attestation and runs annual third-party penetration tests. Decision points run inside your own network and keep answering from cached policy if the control plane is unreachable. See the Trust Center for the full program. Bring your hardest permission problem. Tell us what you are authorizing and where it runs. We come to the call with a model of how Permit would enforce it. Book a demoRead the docs SOC 2 Type II attestedSecurity, Availability, Confidentiality Cloud, hybrid, or on-premDecisions run in your network Every decision loggedExplained, and forwardable to your SIEM Built on open standardsOPA, Cedar, and OPAL Review our security program in the Trust Center