Straiker Defend AI

Defend AI is a runtime layer that sits in front of deployed AI agents — coding copilots (Cursor, Claude Code, GitHub Copilot), productivity agents (Microsoft Copilot, ChatGPT Enterprise), and custom agents built on AWS Bedrock or Azure AI Foundry — and inspects every prompt, reasoning step, and tool call for prompt injection, data exfiltration, and destructive actions. It's deployed as an API/SDK/webhook integration or an inline gateway, not a proxy appliance, and can run in monitoring-only mode or active blocking mode. It's one of three linked products: Discover AI inventories agents and MCP servers, Ascend AI red-teams them pre-deployment, and Defend AI enforces at runtime, with each feeding data back into the others. Source: product page (https://www.straiker.ai/products/defend-ai) and Help Net Security's independent write-up of the March 2026 launch (https://www.helpnetsecurity.com/2026/03/23/straiker-discover-ai/).

Straiker emerged from stealth in March 2025 with $21M from Lightspeed Ventures and Bain Capital Ventures (https://www.straiker.ai/blog/straiker-launches-with-21-million-to-safeguard-ai). In June 2026 it raised a $64M Series A led by Marathon Management Partners with Citi Ventures, Illuminate Financial, and Workday Ventures joining, bringing total funding to $85M (SecurityWeek: https://www.securityweek.com/straiker-raises-64-million-for-ai-security-platform/; SiliconANGLE: https://siliconangle.com/2026/06/29/straiker-lands-64m-defend-enterprise-ai-agents-attack/). Co-founder/CEO Ankur Shah previously ran Prisma Cloud as SVP/GM at Palo Alto Networks; co-founder/CTO Sreenath Kurupati founded Cyberfend (fraud detection, acquired by Akamai) and later led AI/security research at Akamai (https://www.straiker.ai/about). SiliconANGLE reports the company claims run-rate revenue grew more than 15x in under a year off an undisclosed base — a vendor claim, not independently verified.

Straiker's own marketing states Defend AI delivers 98.1% detection accuracy, sub-300ms latency, and 6-21x lower false-positive rates than "frontier model judges" (https://www.straiker.ai/products/defend-ai). Its research arm, STAR Labs, published a July 2026 threat report claiming 36% of successful coding-agent attacks reached remote code execution, 91% of successful productivity-agent attacks ended in silent data exfiltration, and 28.6% of cataloged MCP tools are high-risk by capability alone, drawn from "1,700+ successful exploits" and monitoring of 17,651+ MCP servers (https://www.straiker.ai/report/threat-research-vol-1). We could not find an independent audit, third-party benchmark, or published methodology for any of these figures beyond Straiker's own report — the accuracy/latency numbers in particular have no baseline definition (what counts as a "frontier model judge," what test set was used) on the public site. Separately, STAR Labs' technical research into the China-linked "Villager" AI pentesting tool was substantive enough that independent outlet Hackread cited it directly, which is a genuine (if narrow) signal the research team does real work, distinct from the marketing-benchmark numbers above.

Straiker's own website has no pricing page (straiker.ai/pricing 404s) — this is a contact-sales, enterprise-quote product. AWS Marketplace, where Defend AI and Ascend AI have been listed since July 2025, does publish one concrete figure: a 12-month contract priced at $100,000 for a single line item called "Straiker Platform / Units of platform access" (https://aws.amazon.com/marketplace/pp/prodview-4ey26i3k5epmi). AWS's own AI-generated buyer notes on that same page flag that the listing doesn't define what a "Unit" actually maps to — an agent, a seat, a workload — so that $100K figure is a real anchor point but not something you can size against your own environment without talking to the vendor. For comparison, competing agent-security vendor Zenity publishes an AWS Marketplace rate of $130/resource/month plus $16 per million tokens for runtime protection (https://edgelabs.ai/blog/ai-runtime-security) — a more usage-legible model than Straiker's undefined Unit.

Straiker displays a SOC 2 badge and an ISO/IEC 27001 badge (via auditor Sensiba) on its site, and is a Cloud Security Alliance member (https://www.straiker.ai/about). Its U.S. Data Processing Addendum is public and reasonably detailed, including a security-measures appendix and a subprocessor-notification clause (https://www.straiker.ai/legal/dpa). However, the linked trust center (trust.straiker.ai) returned a certificate error and a 409 on repeated attempts during this review, so we could not independently confirm SOC 2 report type, scope, or date — a buyer will need to request the actual report rather than rely on the badge. Straiker's docs site (docs.straiker.ai) requires a vendor-issued access code and isn't publicly browsable, so integration depth and API design aren't evaluable before a sales conversation.

We could not locate an accessible G2 profile, a Gartner Peer Insights listing, or TrustRadius reviews for Straiker or Defend AI specifically. Customer quotes attributed to named CISOs (Omada Health, Coupa, American Express Global Business Travel, EnterpriseDB, Automation Anywhere) appear only in Straiker's own press releases and website — real named individuals with verifiable titles, which is a stronger signal than an anonymous testimonial, but still vendor-selected and vendor-published, not independently sourced. This is consistent with the AI-runtime-security category generally: a July 2026 independent buyer's guide covering five competing platforms (EdgeLabs, Oligo, Zenity, Astelia, WitnessAI) explicitly notes that several vendors in this space "carry few or zero third-party reviews" and that this is normal for the category rather than a red flag specific to any one vendor (https://edgelabs.ai/blog/ai-runtime-security). Straiker doesn't appear in that comparison at all.