Question 1

What is the Autonomous Workforce and why does it need governance?

Accepted Answer

The Autonomous Workforce refers to fleets of AI agents embedded in core SaaS and cloud platforms that read and write production data, orchestrate tools, and chain workflows across systems. Enterprise AI has shifted from ad-hoc Shadow AI usage to long-lived autonomous agents shipped by major vendors (e.g., Salesforce Agentforce, Bedrock Agents). 40% of enterprise applications are estimated to embed task-specific AI agents by 2026, yet only 21% of organizations maintain a real-time inventory of active agents and just 18% are confident their IAM systems can manage agent identities. This creates a structural time-to-trust gap: enterprises delegate real work faster than they can see, understand, or control it. Arrakis provides Autonomous Workforce Governance — total visibility, policy enforcement, and real-time remediation over these agents.

Question 2

What are the key agentic AI threat vectors — AI worms, RAG poisoning, and cross-agent contagion?

Accepted Answer

Three threat classes define the agentic AI attack surface. AI worms (e.g., the Morris II worm) are prompt-based attacks that self-replicate across GenAI ecosystems by embedding adversarial prompts in emails, shared documents, and tool outputs — because attacks span multiple agents and systems, no individual prompt looks malicious in isolation. RAG poisoning exploits retrieval-augmented generation by injecting crafted passages into knowledge corpora to steer model outputs toward attacker-chosen answers, while indirect prompt injection hides malicious instructions inside web pages, SaaS records, or PDFs that agents retrieve during normal operation. Cross-agent contagion compounds both: a single compromised data source can cascade across an entire fleet of agents through shared tools and corpora. Endpoint-centric AI firewalls that only inspect prompts cannot see these data-plane attacks or reconstruct cross-agent propagation chains — which is why Arrakis governs at the agent, workflow, and SaaS-graph level.

Question 3

What is the MCP attack surface and how does Arrakis govern it?

Accepted Answer

The Model Context Protocol (MCP) enables AI agents to connect to external tools and data sources, but it introduces a new attack surface. Typosquatting on public MCP registries (e.g., registering mcp-postgress-connector to impersonate the legitimate mcp-postgres-connector) exploits the fact that LLMs hallucinate non-existent package names that attackers pre-register. DNS rebinding attacks bypass Same-Origin Policy to reach local MCP servers, with blast radius extending to full remote code execution plus credential theft from all connected services. Arrakis turns the MCP ecosystem into a governed supply chain through its AISPM stage: a dedicated MCP gateway enforces allow-lists and DLP rules before tools are invoked, correlates known CVEs against active MCP configurations, and detects anomalous tool invocation patterns at runtime. This transforms MCP from an ungoverned integration layer into a policy-enforced, auditable tool supply chain.

Question 4

How does Arrakis prevent Financial Denial of Service and excessive agency from AI agents?

Accepted Answer

Agents often inherit broad read-all or administrative permissions and aggressively exercise them — operating strictly within their authorization but in unexpected, high-risk patterns. Excessive agency manifests as bulk data exports at off-hours, unauthorized cross-system data transfers, or agents accessing resources far beyond their intended scope. Financial Denial of Service (Financial DoS) occurs when agents trapped in recursive loops or hallucination-driven retry cycles drive runaway API usage and cloud costs — potentially burning tens of thousands of dollars per day. Arrakis addresses both through its three-stage architecture: AI Observability builds behavioral baselines that flag deviations like sudden 3:00 AM table exports or explosive token consumption, AISPM identifies over-permissioned agents before execution through static analysis, and AIDR enforces hard circuit breakers on token usage and costs at runtime with granular kill-switches that revoke specific access in milliseconds without taking entire platforms offline.

Question 5

What is Arrakis's three-stage governance architecture?

Accepted Answer

Arrakis implements a three-stage governance model treating the agent, its workflow, and the surrounding SaaS graph as the fundamental security object. Stage I — AI Observability: automatic SaaS/cloud agent discovery, ownership tracking, behavioral baseline anomaly detection, and Claude code/API monitoring. Stage II — AI Security Posture Management (AISPM): pre-execution static analysis identifying over-permissioned agents, MCP gateway enforcement with allow-lists and DLP rules, and CVE correlation mapping to OWASP, NIST, and MITRE ATLAS. Stage III — AI Detection and Response (AIDR): Zero-Trust Output treating all agent outputs as untrusted, intent-aware anomaly detection, cross-agent contagion tracking, granular kill-switches revoking access in milliseconds, and circuit breakers for financial DoS prevention.

Question 6

What is Zero-Trust Output and how does Arrakis implement it?

Accepted Answer

Zero-Trust Output is the principle that all agent outputs must be treated as untrusted, regardless of the agent's authorization level. Agents can operate strictly within their permissions but in unexpected, high-risk patterns — exporting entire tables at off-hours, triggering recursive loops that drive runaway API costs (Financial DoS), or producing outputs that have been influenced by poisoned data sources. Arrakis implements Zero-Trust Output through its AIDR stage with intent-aware anomaly detection that evaluates whether agent behavior matches expected patterns, granular kill-switches that can revoke specific access in milliseconds without taking platforms offline, circuit breakers enforcing hard ceilings on token usage and costs, and optional supervisory guardian agents that pre-review high-risk actions.

Question 7

How does Arrakis handle agent identity and non-human identity (NHI) governance?

Accepted Answer

Most organizations lack clear ownership, identity standards, or traceability for autonomous agent actions. Agents increasingly operate as non-human identities (NHI) with credentials and permissions reserved for employees, yet traditional IAM systems are not equipped to govern them. Arrakis models agents as first-class non-human identities through Okta integrations, linking humans, agents, and data stores into a unified identity graph. This provides full traceability of agent actions, ownership tracking, and the ability to enforce identity-aware policies across the agent lifecycle — from discovery through runtime governance.

Question 8

What security frameworks and compliance standards does Arrakis align with?

Accepted Answer

Arrakis aligns with NIST AI RMF (AI Risk Management Framework), MITRE ATLAS (Adversarial Threat Landscape for AI Systems), OWASP Top 10 for LLM Applications, the EU AI Act, Singapore's Model AI Governance Framework, CSA (Cloud Security Alliance) AI guidelines, SOC 2, ISO 42001, and FedRAMP. Built-in AI red-teaming playbooks simulate RAG poisoning and AI worm attacks, while compliance mappings make governance fully auditable. Arrakis's approach — global agent inventory, NHI graphs, behavioral analytics, circuit breakers, and immutable audit trails — provides a practical enterprise instantiation of the emerging global consensus that agents must be governed with the same rigor as human users.

We uncover AI agents' behavior
before anyone thinks to look

Building the future where scaling AI doesn't mean losing visibility

What we build

AI observability

Security posture

Detection & response

What keeps Arrakis on track

Follow the evidence

Fight operational chaos

Enable progress

Say what others avoid saying

Break with tradition

Meet the Founders

Tal Baron

Omer Efrat

Ron Shani

One Team. Total Coverage.

We uncover AI agents' behavior before anyone thinks to look