Straiker

• Platform vendors such as Palo Alto Networks, Microsoft, or the model providers could ship agent testing and runtime protection inside suites enterprises already buy, removing the third-party budget line Straiker depends on, the absorption path the Cato Networks acquisition of Aim Security has already demonstrated.
• The public traction record rests on vendor-published CISO testimonials and one analyst listing, so if no named customer speaks independently, a procurement team could pick a same-asset rival on equal footing despite the larger raise.
• Same-asset independents contest the same enterprise AI buyer, so converting vendor-published testimonials into named reference accounts before the category consolidates further is the open race.
• The 64 million dollar Series A funds a model-training build-out of GPU capacity and pre- and post-training, so if the in-house models do not outperform cheaper bundled guardrails on independent measures, the capital could outrun the commercial proof.
• Capability claims rest on Straiker's own product pages and vendor-reported accuracy figures, so an independent benchmark showing weaker detection than the marketing implies would undercut the positioning.

Straiker treats the enterprise rollout of AI applications and autonomous agents as the asset under attack and sells a platform to secure it. The company frames the buyer as the security team responsible for AI that the business builds and deploys, from customer-facing chatbots to internal agents that hold credentials and take actions. Straiker's platform tests an organization's AI applications against agentic threats including prompt injection, data leakage, and excessive agency.

The threat taxonomy centers on prompt injection and excessive agency, threat types enterprise security teams already track. The problem is the trust boundary an AI agent crosses when it mixes untrusted input with privileged access to enterprise systems.

Independent evidence corroborates the threat class beyond the vendor's framing. CB Insights, in its 2026 AI 100, named model and agent security as a live area of enterprise demand and placed Straiker within it. CVE.org records an allowlist-bypass vulnerability in the Cursor coding agent (CVE-2026-22708) exploitable through prompt injection, and CSO Online reported the publicly available Villager offensive framework, both concrete instances of the agentic attack surface Straiker sells against. [s8, s11, s15, s17]

The Straiker platform runs three modules in a closed loop rather than as single point features. Discover AI inventories the AI agents in an environment, maps their connections, and surfaces misconfigurations. Ascend AI runs continuous red teaming with offensive models that probe multi-stage agent behavior. Defend AI blocks prompt injection, data exfiltration, and agent manipulation in coding, productivity, and custom-built agents at runtime. Straiker describes the three as a closed loop where Discover finds risks, Ascend tests them, and Defend blocks them.

The company attributes the capability to a purpose-built model stack. Straiker says its AI Security Engine uses mixture-of-experts routing and reinforcement learning across fine-tuned models, and that the same engine powers both the offensive testing and the runtime defense, with published runtime figures of 98.1 percent detection accuracy and sub-300-millisecond latency.

The independent evidence is of the research team, not the product. STAR Labs has published in-domain offensive research that independent outlets reported, the NomShub exploit chain in the Cursor coding agent covered by SecurityWeek and the Villager offensive framework covered by CSO Online, which demonstrates offensive craft in Straiker's own threat domain. The cited record includes no public documentation portal or third-party evaluation of the detection figures, so the runtime figures rest on vendor description. [s2, s5, s16, s17]

Straiker competes in enterprise AI security against same-asset independents and the platforms moving to bundle the category. Independent rivals contest the same enterprise AI buyer with discovery, posture, and runtime protection, and Cato Networks' acquisition of Aim Security shows a larger platform buying the capability rather than building it slowly.

Straiker's stated differentiator is the closed loop between offense and defense on one engine. The company positions continuous adversarial testing feeding the same models that enforce runtime protection as a tighter coupling than point tools that do only one side, though that claim rests on vendor description rather than independent evaluation.

The structural question is whether buyers keep paying for a standalone layer. The Cato Networks acquisition of Aim Security shows the capability arriving bundled into a larger platform, which is the bet Straiker asks enterprises to make against. [s11, s2, s5, s19]

Straiker's clearest proof points come from its own launch and a 2026 funding milestone. The company's announcement carries an endorsement from a named CISO at Coupa, and SiliconANGLE names People.ai, Coupa, and DirecTV among its early customers. The customer endorsements, the Coupa CISO quote among them, appear on Straiker's own pages rather than in independent reporting, so they read as vendor-curated references rather than buyers speaking on the record.

A June 2026 Series A is the strongest recent signal. BankInfoSecurity reported a 64 million dollar Series A led by Marathon Management Partners, raising the total to 85 million dollars, and an earlier CB Insights listing characterized 8x growth in six months, a third-party marker of momentum that is not an audited metric or a disclosed customer count.

Commercial scale is otherwise unverified. No named reference customer speaks independently, no public marketplace listing or channel-distribution partnership was observed in the reviewed sources, and no revenue or customer total is disclosed in the public record. The new capital is earmarked for model training, global sales, and support, so the named-customer proof that would move this beyond indirect signals has yet to appear. [s7, s13, s11, s18]

The founding team pairs platform-scale operating experience with a prior security exit. Co-founder and CEO Ankur Shah was previously SVP and general manager of Prisma Cloud at Palo Alto Networks, where the about page says he grew the cloud-security platform from a single module into the company's CNAPP offering, after earlier product roles at RedLock, CipherCloud, and Symantec, and BankInfoSecurity describes him as a former Palo Alto Networks executive.

Co-founder and CTO Sreenath Kurupati brings research and entrepreneurial depth. He was previously co-founder and CEO of Cyberfend, a security startup acquired by Akamai, then served as a vice president leading AI and security research across Akamai's application-security portfolio, after more than a decade at Intel.

The research output and investor signal reinforce the pedigree. Straiker's STAR Labs has produced in-domain offensive research that independent outlets reported, including the NomShub Cursor exploit chain and the Villager offensive framework, and the board includes Bucky Moore of Lightspeed, Gokul Rajaram of Marathon Management Partners, and Enrique Salem of Bain Capital. What is still absent is the category-defining, independently recognized standing that lifts the strongest teams to the top of this dimension. [s6, s16, s17, s18]

Straiker presents the baseline attestation an enterprise security buyer expects. The about page trust strip displays a SOC 2 certified mark, an ISO/IEC 27001 mark issued by Sensiba, membership in the Cloud Security Alliance, and a gold sponsorship of the OWASP AI Exchange, signaling engagement with the AI-security standards bodies. The page names Sensiba as the ISO/IEC 27001 firm but does not state the SOC 2 report scope or its SOC 2 auditor.

The public posture is lighter than the product's sensitivity implies. Straiker's platform inspects an organization's AI traffic and inventories its agents, their permissions, and their runtime behavior, so a security review will likely ask for the underlying SOC 2 report and data-handling terms, and no public trust-center portal with downloadable attestations was observed in the pages reviewed. For a young vendor, that consolidation of assurance evidence is the readiness item most likely to surface in procurement. [s14, s1]

Add analyzed competitors to compare them side by side with Straiker.

Straiker sells to the enterprise security team responsible for the AI applications and autonomous agents the business builds and deploys. SecurityWeek frames the buyer as the organization securing AI chatbots and AI agents, and the product spans the coding agents, productivity agents, and custom-built agents a security team has to govern. The pain is the trust boundary an agent crosses when it mixes untrusted input with privileged access.

The segment widens across three buying moments within one integrated workflow. Discover AI inventories the agents in an environment, Ascend AI tests them offensively, and Defend AI blocks attacks at runtime, so Straiker can enter through a discovery or red-team engagement and expand into runtime protection, where testing and defense share a model engine. That lets one product address an enterprise just starting to map its AI estate and one already running agents in production.

Named demand is recent and concentrated in a launch and early-customer cohort. Straiker's own pages carry endorsements from security leaders at Omada Health, Coupa, American Express Global Business Travel, Enterprise DB, and Automation Anywhere, with its launch page adding People.ai, and SiliconANGLE places it among AI-native security vendors at its March 2025 launch. The open question is whether that early cohort broadens into an independently reported customer base.

Straiker's stated advantage is a closed loop where offense and defense share one model engine. Discover AI maps the agents and surfaces misconfigurations, Ascend AI runs scheduled or on-demand red teaming that quantifies residual risk and tunes guardrails, and Defend AI blocks prompt injection, data exfiltration, and agent manipulation at runtime, with the company describing the three as one system rather than separate point tools.

The engine under the modules blends model types rather than wrapping a single LLM. Straiker says its AI Security Engine uses fine-tuned models, mixture-of-experts routing, and reinforcement learning to reason across the AI and agent stack, powering both the continuous red teaming and the runtime protection, while SiliconANGLE's launch coverage describes the same engine more plainly as a mix of small, fine-tuned models. Defend AI now publishes specific runtime claims of 98.1 percent detection accuracy, 6 to 21 times lower false positives than frontier model judges, and sub-300-millisecond latency.

Independent evidence covers the research team, not the product. STAR Labs has published in-domain offensive research that independent outlets reported, the NomShub Cursor exploit chain via SecurityWeek and the Villager offensive framework via CSO Online, which demonstrates offensive depth in Straiker's own threat domain. No third-party benchmark or independent evaluation of the product's detection accuracy surfaces in fetched sources, so the published accuracy and latency numbers rest on Straiker's own description.

Straiker ran a founder-led, demo-led enterprise launch rather than a self-serve product motion. The company emerged from stealth in March 2025 with two generally available modules and a roster of CISO endorsements, and the launch quotes pair customer references with investor voices from Lightspeed and Bain Capital Ventures.

The traction evidence is real but largely vendor-curated. SecurityWeek and SiliconANGLE confirmed the launch and the seed, and in June 2026 BankInfoSecurity reported a 64 million dollar Series A led by Marathon Management Partners that raised the total to 85 million dollars. The named customer endorsements from security leaders at Omada Health, Coupa, American Express Global Business Travel, Enterprise DB, and Automation Anywhere all appear on Straiker's own pages rather than in independent reporting.

The motion is demo-led into enterprise security teams. Defend AI and the other modules route through a book-a-demo path with no self-serve entry, so the primary visible motion is demo-led direct sales, while a public deal-registration link on the vendor site signals at least some partner motion whose scale the reviewed sources do not document.

Straiker publishes no pricing in fetched sources, so the charged unit and list price stay private. A vendor that hides prices usually targets negotiated enterprise deals, which fits the CISO buyer and the demo-led motion the product pages signal. The absence withholds the budget-anchoring benchmark some peers publish.

The product structure implies more than one possible meter. Discover AI reads as a posture or per-agent scope, Ascend AI as a per-assessment or subscription red-teaming spend, and Defend AI as runtime traffic the engine inspects inline, but which unit Straiker actually charges by is not stated publicly. The closed-loop packaging suggests a platform subscription rather than three separate line items.

The inferable belief is that buyers pay for confident AI deployment rather than tooling features, given the framing around launching AI applications safely. Confirming the unit and whether runtime inspection is metered by volume would require a sales conversation, which the hidden-price posture signals is the intended path.

Straiker is delivered as software the customer integrates, not a service Straiker operates. Defend AI installs through a single hook-based integration, a one-line install via API, SDK, webhook, or AI sensor, with no thick clients, proxies, firewalls, or infrastructure changes required, so the security team wires the guardrail into its own AI traffic and tunes the policy.

The enforcement options fit different parts of the agent stack. Defend AI offers inline blocking, response shaping, and sanitization through an API or SDK, an eBPF sensor, an AI gateway, or a proxy, with privacy-preserving isolated data paths and multilingual coverage, which lets the product sit in front of coding, productivity, and custom-built agents without depending on one platform.

The operational risk profile reads as that of a tool rather than a managed offering. The public pages present Defend and Ascend as customer-integrated software the team installs and tunes itself, and no managed-service tier, support SLA, published uptime, or liability term surfaces in the reviewed pages, so a buyer should read the safety outcome as the customer's responsibility unless a contract says otherwise.

Straiker presents the baseline attestations an enterprise security buyer expects. The about-page certification strip carries badge images for a SOC 2 report and an ISO/IEC 27001 mark issued by Sensiba, alongside Cloud Security Alliance AI corporate membership, an OWASP AI Exchange badge, and an NVIDIA Inception badge. The badge filenames rather than the labels disambiguate the SOC and ISO marks.

The attestations are enterprise-grade but table-stakes rather than a moat. SOC 2 and ISO 27001 ease procurement without blocking a substitute, and no regulatory mandate for this product class appears in the reviewed record, so the marks signal diligence rather than lock-in. No public trust portal with downloadable reports surfaces in fetched pages, so a security review will likely request the underlying SOC 2 report directly.

The product's sensitivity raises the diligence bar beyond the badges. Because Straiker inspects an organization's AI traffic and inventories its agents, their connections, and their runtime behavior, a buyer should resolve data-handling, model-training, and retention terms in a formal review, especially since Straiker says the runtime engine was trained on millions of real-world agent traces, and public sources do not document that corpus's provenance, composition, or any customer-data use.

Straiker positions itself as the security layer across an enterprise's whole AI estate rather than a single point feature. The closed loop spans discovery, offensive testing, and runtime defense in one integrated workflow, with the offensive and defensive halves sharing a model engine, so a security team governs inventory, testing, and enforcement from one product rather than stitching three tools together, which is the platform claim Straiker makes against point vendors.

Outward, the product extends through deployment flexibility rather than a third-party marketplace. The API, SDK, sensor, gateway, and proxy options place enforcement inside environments the buyer already runs, which lets Straiker sell as a neutral layer across model providers and agent frameworks rather than as a feature of one stack. No platform-distribution partnership appears in fetched sources.

The exposure is that the platforms hosting enterprise agents can build the same controls. Cato Networks' acquisition of Aim Security shows a larger platform buying into the category, and Cato can bundle those capabilities into the SASE platform and customer contracts enterprises already hold, so the layer Straiker is building as a standalone is ground larger vendors are positioned to contest by bundling.

Straiker's credibility comes from two repeat security operators. Co-founder and CEO Ankur Shah was previously SVP and general manager of Prisma Cloud at Palo Alto Networks, where the about page says he grew the cloud-security platform from a single module into the company's CNAPP offering and raised its revenue 50x over five years, after earlier roles at RedLock, CipherCloud, and Symantec.

Co-founder and CTO Sreenath Kurupati brings a security exit and research depth. He previously co-founded and led Cyberfend, a security startup acquired by Akamai, then ran AI and security research there, a prior build and exit in an adjacent domain that maps onto Straiker's offensive-and-defensive engine.

The research output and board reinforce the founder signal. Straiker's STAR Labs has produced in-domain offensive research that independent outlets reported, including the NomShub Cursor exploit chain and the Villager offensive framework, and the board pairs the founders with Bucky Moore of Lightspeed, Gokul Rajaram of Marathon Management Partners, and Enrique Salem of Bain Capital. The record still lacks the long, category-defining publication history that lifts the strongest teams in this category.