Evoke Security

• CrowdStrike, which runs the accelerator Evoke reached, and other endpoint vendors can extend detection and response to AI agents. If platform-native agent coverage satisfies enterprise buyers, an independent agent-security layer becomes optional spend.
• No customer or deployment is public. If no named references appear before the next raise, Evoke approaches it on founder credibility without traction proof on a $4 million pre-seed.
• The 2026 accelerator cohort alone included other security startups working the AI-agent problem, such as Above Security and Capsule Security. If the category consolidates before Evoke lands reference customers, it competes on features against better-funded peers.
• The platform installs in minutes and shows no public integration depth or accumulated workflow. Until security workflows come to depend on Evoke, a buyer can drop it cheaply.
• Identity vendors govern AI agents as identities and cloud-security vendors cover AI posture. If those adjacent platforms add agent detection and response, Evoke's combined discovery-and-runtime pitch narrows to a feature.

Evoke Security frames enterprise AI agents as a new class of privileged user that security teams cannot see. The company argues that employees now grant agents access to email, customer data, source code, and production systems, that a single agent can touch dozens of critical systems, and that one compromised agent can exfiltrate data or move across systems without triggering an alert from the existing security stack.

The pain is stated by the company and its investor rather than quantified independently. Crosspoint's Andre Fuetsch says the number one question he hears from CISOs is how to secure agents they cannot see, which supports demand from the buy side but is investor testimony, not third-party measurement. No incident dataset, analyst category, or regulatory driver appears in the fetched record to size the problem beyond the founders' and backers' framing.

The buyer is clear even where the evidence is thin. Evoke sells to enterprise security leaders who are deploying agents faster than they can govern them, and it positions the moment between an agent's decision and its action as the gap legacy tools miss. [s2, s10, s1]

Evoke describes a platform that discovers agents, models their risk, and blocks their actions. The company lists three capabilities, an inventory-and-governance layer that finds every agent, Skill, MCP tool, and data source to remove shadow AI, a posture-and-threat-modeling layer that maps attack paths and flags malicious skills or over-permissioned agents, and a detection layer that watches every agent action and blocks risky behavior such as data exfiltration in real time.

The architecture reaches agents in three places. Evoke covers local agents with a lightweight endpoint sensor, SaaS agents with a browser extension, and production agents with an SDK and proxy, one platform spanning the environments where agents run.

Public verification lags the description. The capability detail lives on the homepage and the funding announcement, and no documentation portal, trial, sandbox, open-source component, or third-party evaluation appears among the fetched pages, so the artifacts a buyer could inspect to test efficacy are not yet public. [s2, s1]

Evoke positions as one control layer for AI agents against both startups and platform incumbents. The pitch is that agents will become the operating system for employees and that security teams have no visibility or control over them, so an independent layer that sees every agent, tool, and action fills the gap. Borrowing the endpoint detection and response label signals the category Evoke wants buyers to place it in.

The startup field around that pitch is already crowded. The same CrowdStrike, AWS, and NVIDIA accelerator cohort included Above Security and Capsule Security among other AI- and cloud-native security startups, and more companies are forming around the same emerging budget.

The stronger competitive pressure is absorption. Endpoint vendors, CrowdStrike among them, and identity platforms that already govern non-human accounts can extend into agent detection and response, and CrowdStrike judging the accelerator that named Evoke a finalist shows how close the incumbent sits to the same problem. [s3, s5, s2]

No demand evidence is public. The fetched sources name no customers, design partners, case studies, or commercial partnerships, and the accelerator post closes by inviting security teams to start a proof of value rather than pointing to deployed references. Months from its funding announcement, that absence fits the stage while remaining the central gap in the record.

The founders front the effort and the team is early. The careers page lists no open roles and directs interested people to follow the company on LinkedIn, and no go-to-market hires appear in the public materials, which is consistent with founder-led selling at the pre-seed stage.

Indirect signals stand in for traction. Crosspoint Capital, a cybersecurity-focused fund, led the pre-seed with Red Cell Partners, the CrowdStrike, AWS, and NVIDIA accelerator selected Evoke as one of six finalists from nearly 1,000 applicants, and Evoke joined the NVIDIA Inception program. Those endorsements open enterprise doors, and none of them is a paying customer. [s3, s5, s11, s2, s1]

The founders' backgrounds fit the problem. Co-founder and CEO Jason Rebholz was chief information security officer at Corvus Insurance and is described on the site as an incident response leader, and co-founder and CTO Jeff Chan is described as a former CTO and security engineering leader. Incident response and security engineering are close prior art for watching and governing AI agents.

Investor testimony reinforces the pedigree. Red Cell's George Barnes, a former deputy director of the NSA, says the founding team investigated nation-state breaches at Mandiant, built security engineering systems, and led enterprise security teams. That statement comes from a backer rather than an independent record, and it speaks to the team collectively rather than documenting each founder.

Bench depth beyond the two founders is not yet assessable. The fetched sources name no other executives and list no open roles, so the credibility comes from the founding pair and the investors who vouch for them. [s1, s9, s11, s2]

Trust collateral is early and mostly gated. Evoke runs a trust center at a dedicated subdomain, but the public page shows only an access-request form, so no certification, audit report, or security documentation is publicly inspectable. No public attestation such as SOC 2 or ISO 27001 was found by a probe of the trust subdomain, the homepage and platform footers, and the site security paths as of July 8, 2026.

The product's position raises the bar buyers will apply. Evoke asks to place an endpoint sensor, a browser extension, and an SDK inside enterprise systems and to block agent actions in real time, so procurement reviews will probe data handling, availability, and failure modes that no public material yet addresses.

Company age lengthens every one of those reviews. A pre-seed vendor inserting itself into the path of agent actions will face security and legal scrutiny that founder credibility helps with but does not resolve, and the gated trust center is the beginning of the answer rather than the whole of it. [s7, s2]

Add analyzed competitors to compare them side by side with Evoke Security.

Evoke targets enterprise security teams at organizations deploying AI agents across employee, SaaS, and production environments. The three deployment surfaces name where the company expects agents to live, local agents on endpoints, SaaS agents inside the browser, and production agents behind an SDK and proxy.

Segmentation below that is undisclosed. The fetched sources name no verticals, company sizes, or geographies, and with no customers public, the served segments are inferred from the product's reach rather than shown by deployments.

The buyer is a security leader who has lost visibility into agent activity. Crosspoint frames the demand as CISOs asking how to secure agents they cannot see, which locates Evoke in the enterprise security budget even though the specific buyers remain unnamed.

Evoke's claimed capabilities are concrete and its architecture is specific. The company discovers every agent, Skill, MCP tool, and data source, maps attack paths and flags over-permissioned agents and malicious Skills or MCP tools, and watches every agent action to block risky behavior such as data exfiltration in real time, reaching agents through an endpoint sensor, a browser extension, and an SDK and proxy.

The claimed advantage is coverage and runtime control rather than a data asset. Evoke argues that existing security tools miss autonomous agent activity, that one layer spanning local, SaaS, and production agents closes that gap, and that blocking at the moment of action is what those tools cannot do.

Demonstrated efficacy is not yet public. No documentation portal, benchmark, trial, open-source component, or third-party evaluation appears in the fetched sources, so the capability rests on the company's own description and the founders' track record rather than on inspectable proof.

Evoke sells to enterprises through a proof-of-value motion with no self-service path. The accelerator post invites security teams to start a proof of value, the careers page lists no open roles and points to LinkedIn, and the reviewed pages name no sales hires, so with only the founders publicly named the selling is likely founder-led at the pre-seed stage.

Distribution so far runs through credibility rather than channel. Evoke earned a finalist spot among six of nearly 1,000 applicants in the CrowdStrike, AWS, and NVIDIA accelerator, joined the NVIDIA Inception program, and publishes practitioner content on securing agents, a motion that reaches buyers and ecosystem partners without a quota-carrying team.

No marketplace listing, reseller, or managed-service motion appears yet. For a company months past its pre-seed, that is the expected posture, and it leaves the accelerator and investor networks doing the work customer logos would otherwise do.

Evoke publishes no pricing, and no pricing page appears among the fetched pages, which signals negotiated enterprise deals and a sales-assisted motion. For a pre-reference vendor that posture is expected, and it leaves buyers no way to size the spend without a conversation.

The charging unit will say the most once it is set. Pricing per monitored agent would match how buyers count the problem, pricing per environment or seat would track deployment breadth, and nothing public commits Evoke to either, so the model that reveals what the company thinks buyers pay for is still undisclosed.

Evoke deploys in minutes and claims no disruption to employee workflows. The endpoint sensor, browser extension, and SDK-and-proxy design lets the platform reach agents wherever they run, one deployment spanning local, SaaS, and production environments.

Inline blocking sets an operational bar the public record does not address. Stopping agent actions before they execute places Evoke in the path of business workflows, so latency, availability, and failure-mode questions will dominate technical evaluations, and no service-level commitment, status page, or deployment documentation appears in the reviewed sources.

The connect-in-minutes design that eases adoption also lowers the cost of leaving. With no public integration depth or learned workflow tying a customer in, a buyer that removes the sensor and extension loses coverage without unwinding a deep deployment.

Trust collateral trails the access the product asks for. Evoke runs a trust center on a dedicated subdomain, but its contents sit behind an access-request form, and a rendered check of the portal returned only that form, so no certification, audit report, or security page is publicly inspectable and none was found by a probe of the trust surfaces as of July 2026. These are the beginnings of a trust program rather than a moat.

The product's position raises the bar buyers will apply. A vendor that places a sensor, a browser extension, and an SDK inside enterprise systems and blocks agent actions in real time will face procurement reviews on data handling, availability, and failure modes that no public material yet answers.

Closing that gap is cheaper than losing a procurement cycle to it. Evoke sells auditable visibility and control over agents while publishing little about its own security posture, and company age will lengthen every enterprise review until the trust center opens.

Breadth across agent environments is Evoke's platform claim. One layer spanning local, SaaS, and production agents is the coverage argument, and the endpoint sensor, browser extension, and SDK are the mechanism that reaches agents wherever they run.

The ecosystem ties so far are accelerator, program, and investor relationships rather than technical integrations. Evoke sits in the CrowdStrike, AWS, and NVIDIA accelerator and the NVIDIA Inception program, which place it near the hyperscaler and endpoint platforms that could become partners or absorbers.

No partner program, marketplace listing, or public integration catalog appears yet. The platform depth that would let other tools build on Evoke, or lock a buyer in, is not visible in the fetched record.

The founding pair maps to the problem from two directions. CEO Jason Rebholz was chief information security officer at Corvus Insurance and is described as an incident response leader, and CTO Jeff Chan is described as a former CTO and security engineering leader, backgrounds that fit watching and governing agents at runtime.

Investor testimony carries part of the credibility. Red Cell's George Barnes, a former deputy director of the NSA, says the founding team investigated nation-state breaches at Mandiant and led enterprise security teams, a statement from a backer rather than an independent record.

Bench depth beyond the founders is undisclosed. The fetched sources name no other executives and list no open roles, so the founding pair and the investors vouching for them carry the team story.