The Agentic Horizon | SailPoint

Summary of technological concepts & product features

1) Why AI agents change identity security

• The speakers argue enterprises are shifting from an automated model to an autonomous enterprise, where AI agents reason, decide, and act at machine/AI speed (24/7).
• Traditional human-centric identity governance—joiner/mover/leaver processes on quarterly/weekly cadences—is too slow for agent-driven access.
• Key risk theme: enterprises will see an increasing number of non-human identities (agents/service accounts) operating “in the shadows,” while APIs become a top attack vector.

2) Core promise: “SailPoint Agentic Fabric”

SailPoint positions Agentic Fabric as an identity-powered, end-to-end control plane for AI agent identity security—designed as a unified alternative to piecemeal discovery and governance solutions.

The “three pillars” of Agentic Fabric

1. Discovery + Unified Registry

   • Detects and inventories agents across multiple surfaces, then correlates them to human identities.
   • Discovery sources mentioned include:
     • Cloud agent platforms (e.g., AWS Bedrock Agent Core, Microsoft, Google Cloud, etc.)
     • Enterprise apps (e.g., Salesforce, SAP, Oracle, Workday, etc.)
     • Endpoints (coding/agent tools such as “Claude/Code, Cursor, OpenAI tools”, etc.)
     • Browser-based agents
     • Network/gateway monitoring (“gateway”)
   • Key capability: automatically correlate agent identities to human ownership/context using SailPoint’s identity context—contrasted with competitors that rely on shallow mappings (e.g., directory groups or crowdsourcing).

2. Real-time Governance + Compliance/Audit

   • Lifecycle management for agents, including access requests and handling for agents/credentials/service accounts.
   • Just-in-time / real-time authorization rather than standing privilege.
   • Auditability for agents:
     • Evidence generation
     • Support for multiple regulatory/audit frameworks
     • “One-click evidence generation”
   • Claim: agent auditing frameworks consolidate into about 9–10 controls, with “agent audit support out of the box.”

3. Authorize, Protect, Respond (risk + runtime controls)

   • Behavioral monitoring to detect drift and malicious or out-of-mission behavior.
   • Prompt injection / prompt security:
     • Scan prompts
     • Block prompts that include risky or sensitive-data content
   • Shadow AI remediation:
     • Automatically block unsanctioned agents/tools
     • Remediate risk without manual firefighting
   • Real-time response with blast-radius containment, including integration with security tooling (e.g., SIEM/SOC).
   • Human-in-the-loop certification campaigns when new access/changes require owner attestation.

3) “Six foundational principles” for trustworthy agent governance

Across the keynote/tech sections, the platform is described as built around agent-security principles:

• Immutable ownership / birthright: each agent has a designated accountable human owner that agents can’t change.
• Real-time ledger / traceability: maintain an unchangeable record of agent activities.
• Intent-based continuous risk evaluation: continuously reassess risk as agents operate.
• Real-time authorization via privilege: emphasize just-in-time access and avoid standing privilege.
• Cryptographic verification for agent-to-agent / agent-to-tool interactions to prevent abuse via token/credential passing.
• Agent “moral code” / guardrails: ethical/policy guardrails to prevent agents from acting outside allowed behavior.

A real-world example was cited: an Air Canada AI chatbot fabricated a refund policy, leading to legal/accountability issues—used to emphasize why ownership immutability matters.

4) Live demo highlights (how it works in practice)

• Setup in minutes:
  • Identity provider connection
  • Sensor deployment
  • SIEM connection
  • Cloud integration (example showed an AWS agent core deployment)
• Dashboard/registry shows:
  • Agent types (enterprise/platform agents, endpoint agents, browser agents, etc.)
  • Ownership mapping and risk scores
  • Tool counts, MCP servers
  • An “identity graph” linking humans → agents → machines/apps → data/entitlements
• Automatic remediation workflows:
  • A “shadow agent” identified as unsanctioned is auto-remediated (e.g., disabled/blocked access).
  • An endpoint/agent drift example:
    • If an agent connects to an unsanctioned MCP server, risk spikes
    • Policy removes the bad connection and risk returns to green without manual intervention.
• Succession planning / attestations:
  • Long-lived “HR-style” agents get periodic ownership review and reassignment when owners leave, maintaining an auditable chain of custody.

5) Risk management features emphasized

• Risk center
  • Shows risk distribution and trajectories over time
  • Example claim: ~50% reduction in high-risk agents over a week
  • Drivers cited: shadow AI remediation, just-in-time access removing standing privilege, and human review/oversight
• ABM (Agent Behavior Monitoring)
  • Establishes a baseline of allowed tool usage patterns and operating times
  • Detects drift proactively and blocks/mitigates before harmful behavior completes
• Prompt security center
  • Scans prompts (example: thousands scanned in a week)
  • Blocks prompt misuse (e.g., sensitive data in prompts, risky document uploads)

6) Compliance / audit workflow

• One-click evidence generation for frameworks such as:
  • SOC 2, NIST, GDPR, and the EU AI Act (mentioned as built-in support)
• Custom framework support:
  • Teams can define/select controls and generate audit reports
• Scheduled report generation (weekly/monthly/quarterly) to prepare evidence ahead of audits

7) Packaging / adoption guidance

• New simplified Agentic packages announced:
  • Agentic Business
  • Agentic Business Plus
• Adoption narrative:
  • Start with discovery and least privilege
  • Progress toward zero standing privilege and more advanced just-in-time controls
• Additional discussed topics:
  • Investment protection plan for customers already on certain products
  • Integration/compatibility across deployment scenarios (cloud vs on-prem identity governance), including IdentityIQ/IIQ, identity security cloud, and legacy/third-party governance

---

Main speakers / sources (as presented in the subtitles)

• Kelly — CISO (customer story using SailPoint to protect identities/apps)
• Wendy Woo — Chief Marketing Officer, SailPoint (event host)
• Matt Mills — President, SailPoint (keynote)
• Rex Stenton — Senior Managing Director, Accenture (partner discussion)
• Chandra — Executive Vice President & Chief Technology Officer, SailPoint (Agentic Fabric technology)
• Levent Bess — Chief Product Officer, SailPoint (product/capability overview)
• Kirby — Product team (live demo walkthrough)
• Y — Product manager (risk, ABM, prompt security demo segment)
• John — mentioned briefly (likely an executive/producer voice in the product story)
• Meredith Banchard — Chief Customer Officer, SailPoint (go-to-market / playbook / roadmap)
• Additional implied / referenced sources:
  • Gartner prediction
  • Air Canada refund policy tribunal
  • Common frameworks/regulations referenced: DORA, EU AI Act, SOC 2, NIST, GDPR