Akeyless Agentic Runtime Authority
by Akeyless
Intent-based runtime access control for AI agents: credentialless access, per-action policy checks and a session kill switch
Akeyless Agentic Runtime Authority is an identity control layer for AI agents that brokers their access to databases, clouds and servers through the Akeyless Gateway and checks each requested action against policy before it runs. It is aimed at platform and security teams putting coding and business agents into production against real systems.
Akeyless Agentic Runtime Authority (ARA) is a feature of the Akeyless Identity Security Platform, which also covers secrets management, certificate lifecycle management, privileged access and password management. It entered a private beta with design partners in March 2026 and became generally available to all Akeyless customers on September 9, 2026. ARA works in two layers. SecretlessAI keeps credentials out of the agent: requests go through the Akeyless Gateway, which provisions a short-lived identity on the target system, including legacy and on-premises systems, so the agent never holds a long-lived secret. Runtime Authority then governs what the agent does with that access. Administrators attach input rules, which constrain the queries, prompts or commands an agent can send and deny violations before they reach the target, and output rules, which filter or redact what comes back. SQL targets default to read-only with no multi-statement queries. An optional AI Quorum mode evaluates each action against several configured LLMs and blocks it if any one flags a violation. Supported targets in the documentation include MySQL, PostgreSQL, MSSQL, Oracle, Snowflake, Redshift, MongoDB, Redis and Cassandra, plus AWS, GCP, Azure, Kubernetes, GitHub, SSH, Windows and LDAP. Agents connect through an MCP server or the CLI, with templates for Claude Desktop, Cursor and GitHub Copilot; the GA release added Claude Enterprise, OpenAI Codex and Amazon Bedrock AgentCore. Every session is logged to the triggering human, application or agent, can be terminated from an Agentic Access Dashboard, and can be forwarded to Splunk, Datadog or Microsoft Sentinel. Akeyless competes in secrets management with HashiCorp Vault and the cloud providers' secret managers.
A platform security or IAM team that is letting AI agents query production databases and run cloud or shell commands, and needs per-action control instead of handing agents standing credentials.
Agents get short-lived, brokered access with each action checked against intent and policy, a logged audit trail and a way to kill a session immediately.
At a Glance
- Category
- Governance & Security
- Pricing
- Freemium, Usage-based, Contact for pricing
- Target Market
- CISOs, CTOs, Platform Engineering Teams, IAM and Privileged Access Teams, Enterprise Developers
- Deployment
- Cloud-first, Hybrid, API-based
Key Features
- ✓SecretlessAI credential brokering
Routes agent requests through the Akeyless Gateway and provisions short-lived identities on the target, so agents never hold long-lived secrets.
- ✓Input and output rules
Input rules deny disallowed queries or commands before they reach the target; output rules filter or redact data returned to the agent.
- ✓AI Quorum evaluation
Optionally checks each action against multiple configured LLMs and blocks it if any model flags a policy violation, reducing single-model misjudgments.
- ✓Agentic Access Dashboard and kill switch
Shows live agent sessions, explains why a session was blocked and lets an administrator terminate a running session immediately.
- ✓MCP server and CLI access
Exposes list-secrets, query-db and service-execute tools over MCP, with config templates for Claude Desktop, Cursor and GitHub Copilot.
- ✓Attributed audit and SIEM forwarding
Logs every action to the human, application or agent that triggered it and forwards enriched events to Splunk, Datadog or Microsoft Sentinel.
Capabilities
Use Cases
- •Safe database access for coding agents
Developers let Cursor or Copilot query a Snowflake or PostgreSQL replica through the MCP server under default read-only, single-statement rules.
- •Task-scoped business agents
An agent told to summarize sales data is blocked from deleting a database or exfiltrating bulk records, even though its underlying permissions would allow it.
- •Incident containment for misbehaving agents
When an agent starts issuing unexpected commands, security sees the blocked actions in the dashboard and terminates the session at once.
- •Audit-ready agent operations
Compliance teams trace every agent action against AWS, Kubernetes or SSH targets back to a named human or application in the SIEM.
Ideal For
Best For
- ✓Giving coding agents such as Cursor, GitHub Copilot or Claude read-only database access without putting credentials in the agent's config
- ✓Enforcing task scope on agents, for example blocking a data-summary agent from deleting tables even when its role technically allows it
- ✓Organizations already running Akeyless for secrets management or privileged access that want to extend it to agents
- ✓Security teams that need agent actions attributed to a human or application and streamed into Splunk, Datadog or Sentinel
- ✓Brokering agent access to legacy and on-premises systems through a gateway rather than exposing them directly
Not Ideal For
- ✗Teams that only need shadow AI discovery or prompt monitoring for SaaS assistants; ARA governs what agents do against infrastructure targets, not what employees type into chat tools
- ✗Organizations not prepared to run the Akeyless Gateway and enable AI Insights with one or more LLMs, which the documentation lists as prerequisites
- ✗Buyers who need published pricing; the pricing page does not list agent features or any per-agent unit
Deployment
Market Analysis
Pros
- ✓Removes standing credentials from agents by brokering short-lived access through a gateway
- ✓Broad target coverage in the docs, from Snowflake and Oracle to Kubernetes, SSH and Active Directory
- ✓Session-level kill switch and attribution to the triggering human or application, with SIEM forwarding
Cons
- ✗Requires an Akeyless Gateway plus AI Insights configured with one or more LLMs, and works only against secrets already defined in Akeyless
- ✗The documentation does not state the latency an LLM policy check adds to each action, or describe a monitor-only mode
- ✗No independent user reviews or Hacker News discussion of the agent features turned up; coverage so far is the vendor's release and trade-press write-ups of it
- ✗Pricing for agent features is unpublished, and Akeyless's own pages give conflicting volume figures (220 billion versus 20 billion monthly identity interactions)
Pricing
Free
$0
- ✓Secrets Management: 5 clients, 500 static, 5 dynamic and 5 rotated secrets
- ✓1 gateway cluster
- ✓3 days of audit log retention
- ✓Agent features not listed on the pricing page
Enterprise
Contact for pricing
- ✓Custom quotas
- ✓Pure SaaS or Hybrid SaaS with on-premises gateways and Zero-Knowledge encryption
- ✓SAML/OIDC/LDAP authentication, log and event forwarding, session recording
Akeyless publishes a Free plan and a custom-quoted Enterprise plan, metered by clients, connectors, certificates, transactions and users depending on the product. The pricing page, checked November 16, 2026, does not mention Agentic Runtime Authority or SecretlessAI or any per-agent unit, so agent pricing has to come from sales. Log forwarding and SSO via SAML or OIDC are Enterprise-only.
Security & Compliance
Sources
This page was written from 6 sources, 5 on domains other than akeyless.io.
- 1.prnewswire.com — akeyless announces general availability of agentic runtime a
- 2.docs.akeyless.io — agentic runtime authority
- 3.akeyless.io — pricingvendor
- 4.helpnetsecurity.com — helpnetsecurity.com
- 5.techcrunch.com — akeyless secures a cash infusion to help companies manage th
- 6.hn.algolia.com — search
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Abnormal AI Governance
Shadow AI discovery, AI agent inventory and AI spend tracking built from email, identity and OAuth signals
Darktrace / SECURE AI
Behavioral monitoring of enterprise AI use: shadow AI blocking, prompt risk analysis and AI agent identity mapping
Pillar Security
AI security platform for discovering, red-teaming and guarding AI agents, MCP servers and coding agents, being acquired by Vanta
Bay
AI agent security platform that discovers every agent and MCP server and blocks risky agent actions at runtime