Straiker is an agentic AI security company. It gives security teams a view of every AI agent running in the enterprise, tests how those agents can be attacked or go wrong, and stops harmful behaviour while they run, so organisations can put agents to work with access to code, tools and business systems without losing control of them.
Its buyers are security teams in companies that have moved past AI pilots, where coding assistants, custom agents and agent builder platforms are already connected to real data and tools.
At A Glance
| Vendor | Straiker |
|---|---|
| Category | Agentic AI security |
| Identities covered | AI agents, MCP servers, agent tools and permissions |
| Deployment | SaaS or on premises |
| Agentic AI | Agent discovery, red teaming, runtime guardrails, kill switch |
| Customers | Omada Health, Coupa, EnterpriseDB, Automation Anywhere, Emergent |
| Certifications | SOC 2, ISO/IEC 27001 |
| Market classification | Agentic AI Identity |
What Straiker Does
AI agents now read code, call tools and reach into enterprise systems on their own, which gives attackers new ways in through prompts, poisoned tools and over-broad permissions. Straiker’s answer rests on three linked steps: find the agentic activity, test the real attack paths, and defend the agents at runtime. What testing finds feeds the runtime policies, so the controls on live agents reflect the weaknesses actually found in them rather than a generic rule set. When something does go wrong, the response can be immediate and contained to the agents involved.
Products
Discover AI
Most security teams cannot say how many agents are running or what each one can reach. Discover AI answers that by cataloguing agents, tools, MCP servers, Claude Skills and integrations in real time, then assessing each agent’s configuration, permissions and blast radius. Over-permissioned agents and risky MCP connections are flagged, and risky MCP or tool access can be blocked or revoked.
Ascend AI
Ascend AI is for teams shipping agentic applications who want to find weaknesses before an attacker does. It attacks agents the way a real adversary would, looking for prompt injection, MCP tool misuse, agentic exploits, data leakage and exfiltration, including PII, PCI and HIPAA data. Tests run continuously, on a schedule or on demand across development, staging and production, and the CI/CD integration triggers a new assessment when a model, prompt or configuration changes.
Defend AI
Defend AI places guardrails around live agents, inspecting each prompt, reasoning step and tool call. It blocks direct and indirect prompt injection, jailbreaks, data exfiltration, agent and tool manipulation, malicious or vulnerable MCP servers, toxic output and policy violations. Teams choose between Detect and Protect modes and can write their own policies. Straiker reports 98.1% detection accuracy at under 300ms latency.
AI Agent Kill Switch
When an agent misbehaves, the kill switch cuts off its tool access, suspends its sessions and isolates it, for one agent or many at once, while keeping the evidence for investigation.
How It Works
Defend AI runs as SaaS, or on premises for regulated and sovereign environments. Setup is a one-line install, with no thick clients, proxies, firewalls or infrastructure changes.
| Component | What it does |
|---|---|
| Connection methods | Defend AI connects through an API, SDK, webhook or AI sensor; Ascend AI works from an API, logs, an SDK or AI sensors. |
| Discovery sources | Discover AI collects from endpoints, APIs, gateways, SaaS applications and inference hooks. |
| Runtime inspection | Tracing follows each interaction across the user, the model, the tool chain and agent-to-agent calls. |
| Alerting and evidence | Alerts and logs go out over Slack, email or webhook, alongside dashboards and audit logs. Audit evidence is mapped to OWASP, MITRE ATLAS, NIST AI RMF and the EU AI Act. |
Identities And Systems Covered
| AI agents | Every agent operating across the environment, including coding agents and custom-built agents |
|---|---|
| MCP servers and skills | MCP servers and Claude Skills connected to agents |
| Permissions | Agent configurations, permissions and blast radius |
| Agent builder platforms | LangGraph, CrewAI, AutoGen, Vercel, AWS Bedrock AgentCore, Azure AI Foundry, Microsoft Copilot Studio, Google Vertex AI, Databricks, Snowflake |
| Consumer and coding agents | Anthropic Claude, OpenAI Codex, ChatGPT Enterprise, GitHub Copilot, Cursor, Microsoft Copilot, Perplexity |
Agentic AI Capabilities
Threat Coverage
Red teaming and runtime protection share one threat model. Ascend AI tests for goal hijacking and inter-agent manipulation as well as injection and tool misuse, and Defend AI also stops identity abuse, memory poisoning and resource exploitation. Coverage is organised by Straiker’s STAR framework, which spans user, identity, model, data, agent and LAVA risks, together with the OWASP Top 10. Posture monitoring then tracks agents and agentic workflows over time, so a misconfigured or unsafe integration is caught when it appears rather than at the next review.
Company Background
| Founders | Ankur Shah and Sreenath Kurupati |
|---|---|
| Headquarters | Mountain View, California, United States, with an office in Bengaluru, India, and remote teams |
| Leadership | Ankur Shah, Co-Founder and CEO; Sreenath Kurupati, Co-Founder and CTO |
| Funding | A reported $21 million in initial funding from Lightspeed Ventures and Bain Capital Ventures |
| Other investors | Citi Ventures and Workday Ventures |
| Recognition | Included in the Gartner Hype Cycle as a sample vendor for agentic AI security |
Note: This overview is compiled from Straiker’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.