Comp AI Funding Reaches $34 Million, but Continuous Security Is the Harder Bet

Comp AI raised a $34 million Series A, giving the young compliance startup fresh capital to pursue a much larger cybersecurity role. The Comp AI funding round supports an expansion from audit preparation into continuous monitoring, control validation, and security testing.

That shift matters more than the financing alone. Compliance platforms traditionally collect evidence, track controls, and prepare organizations for periodic assessments. Comp AI wants its agents to perform parts of the underlying work and react as risks change.

This ambition puts the company against established compliance platforms such as Vanta and Drata. It also creates a harder contest between automated execution and the human judgment required for trustworthy security decisions.

The round was announced on September 17, 2026. Roo Capital and Grand Ventures led it, according to the company’s funding announcement. Comp AI says it has raised $37.5 million in total, although some reports calculate the total at $36.6 million.

The discrepancy likely reflects different treatments of earlier financing. It does not change the central story. Investors are backing Comp AI’s attempt to turn compliance software into an always-running security layer.

Comp AI Funding Backs a Move Beyond Audit Preparation

The $34 million round funds a product expansion, not simply a larger version of Comp AI’s current compliance service.

Comp AI was founded in January 2025 by Lewis Carhart, Claudio Fuentes, and Mariano Fuentes. Carhart serves as chief executive, Claudio Fuentes as chief operating officer, and Mariano Fuentes as chief technology officer.

The founders previously worked together on LeapAI, an AI workflow product that reportedly reached more than one million users. They eventually closed it after failing to find a sufficiently durable use case.

Their experience pursuing larger customers exposed another problem. Enterprise buyers often demand security documentation before signing a contract, yet preparing for that review can consume months of attention.

Comp AI began by automating parts of that process. Its software connects to a customer’s systems and documents, gathers evidence, drafts policies, builds risk registers, and monitors compliance controls.

Governance, risk, and compliance software, commonly called GRC, helps organizations document requirements and manage evidence for audits. Comp AI applies language models and software agents to tasks that employees or consultants previously handled manually.

The platform supports frameworks including SOC 2 and ISO 27001. SOC 2 is an independent assessment of controls related to areas such as security, availability, and confidentiality.

The company says its product also handles customer onboarding, vendor security assessments, and policy generation. These functions can shorten the administrative work surrounding an audit, but they do not eliminate the audit itself.

Carhart told technology reporting that the platform does not replace independent review. Humans still approve policies, support controls, and oversee automated workflows.

That limitation is important because compliance evidence needs more than speed. It must accurately represent how an organization operates and whether its controls work as described.

Comp AI says it now serves more than 1,000 customers. It also reports annual recurring revenue growth of 15 times year over year. Those figures come from the company and have not been independently audited in public.

Reported customers include Dub, OpenCode, Corgi, Inference, and Primer. The company also says it works with organizations in healthcare, financial services, higher education, and professional services.

Those sectors have different obligations, data risks, and approval requirements. Serving them requires more than generating standard policies or matching evidence to a checklist.

The new capital will support hiring across engineering, product, operations, sales, marketing, and customer success. Comp AI has offices in the Miami area and New York.

More importantly, the company plans to build continuous monitoring and control validation across applications and infrastructure. It also wants agents to conduct security testing and help respond when risks change.

That product direction turns the Comp AI funding story into a test of scope. Audit preparation is structured and document-heavy. Continuous cybersecurity deals with changing systems, ambiguous signals, and actions that can affect production environments.

The jump is attractive because compliance already provides an entry point into customer systems. Once connected, the platform can observe controls, permissions, evidence, and operational changes.

However, access does not automatically create reliable security judgment. Comp AI must prove that its agents can interpret those changes without producing excessive alerts or unsafe actions.

The financing gives the company time and staff to pursue that proof. It does not provide the proof itself.

Why Point-in-Time Compliance Is Under Pressure

Comp AI’s strongest argument is that a completed audit cannot describe every system change that follows it.

A compliance audit evaluates controls during a defined period. The resulting report can support sales, procurement, risk reviews, and customer assurance.

Yet software environments keep changing after auditors finish their work. Teams deploy code, add vendors, modify permissions, and introduce new data flows.

AI agents accelerate that rate of change. An agent can access customer information, call internal tools, update records, or generate code across several systems.

Carhart offered a clear example. A company might complete a SOC 2 audit and deploy a new AI agent two weeks later. That agent might receive access to customer data or change internal permissions.

The audit does not immediately become invalid. It simply was not designed to report every later change in real time.

This gap creates the opening behind the Comp AI funding round. Comp AI wants to monitor the environment between formal assessments instead of treating audit readiness as a seasonal project.

Continuous compliance generally means collecting evidence and checking controls throughout the year. Continuous security goes further by testing whether protections still work and responding to new risks.

Comp AI says agentic AI makes that second step more practical. Agentic systems can plan tasks, use connected tools, and take actions within defined boundaries.

A monitoring agent might notice that a critical employee account lost multifactor authentication. Another could detect an unapproved cloud resource or evidence that has become stale.

A more advanced system could initiate a test, assemble supporting evidence, and recommend a response. It might also open a remediation task or request human approval.

This approach promises a tighter link between compliance records and operational reality. Instead of asking whether a control existed during an audit, customers could ask whether it works now.

Claudio Fuentes described historical compliance as an approximation of security because much of its work occurred manually and periodically. The company’s thesis is that agents remove part of that timing constraint.

That claim addresses a recognized weakness in checkbox-driven compliance. Passing an audit does not guarantee that an organization can prevent, detect, or contain every attack.

A well-maintained compliance program can still improve security discipline. It creates ownership, evidence, repeatable procedures, and an external review point.

The problem appears when certification becomes the destination rather than one verification mechanism. Teams can focus on producing evidence while overlooking whether controls remain effective after collection.

Comp AI’s proposed system attempts to keep that evidence connected to live conditions. This direction also reflects a broader convergence among compliance automation, cloud security, and application security products.

The convergence increases the pressure on established GRC vendors. Customers may begin expecting platforms to validate controls instead of only recording their status.

Vanta and Drata already offer automated evidence collection and continuous monitoring features. They also have larger installed bases, established auditor relationships, and mature integrations.

Comp AI therefore cannot win by claiming that continuous monitoring is entirely new. Its differentiation depends on how much work its agents can safely perform.

Investors believe the timing favors that approach. Roo Capital principal Varun Sridhar argued that businesses cannot treat compliance as an annual exercise while AI changes security environments quickly.

Grand Ventures general partner Nathan Owen cited portfolio companies that moved from other compliance platforms to Comp AI. That statement is an investor endorsement, not an independent market study.

Still, it points to a practical source of demand. Smaller companies want compliance support without building large internal teams, while larger buyers expect deeper operational proof.

That tension creates a market for software that does more than maintain checklists. It also raises the consequences of automation errors.

Comp AI Versus the Compliance Systems of Record

The primary contest is between agents that perform compliance work and platforms that mainly organize evidence for people.

The first generation of compliance automation reduced dependence on spreadsheets, shared folders, and repeated evidence requests. It connected business systems to standardized control frameworks.

That model made compliance easier to manage. However, employees and consultants still completed much of the interpretation, remediation, and approval work.

Comp AI says agents can close more of that loop. The system is intended to understand organizational context, identify required work, perform bounded tasks, and verify outcomes.

Organizational context includes a company’s systems, policies, employees, vendors, and risk profile. Without it, an agent might generate generic documents that do not match actual operations.

Comp AI reads connected systems and existing materials to build that context. It can then draft policies or risk records that reflect the customer’s environment, according to the company.

This mechanism helps explain the attraction of AI-powered compliance software. Language models can work across documents, questionnaires, control descriptions, and unstructured evidence.

They can also translate a technical change into compliance language. For example, a new cloud permission may affect several controls across different frameworks.

Traditional rules can detect the changed permission. An AI system can potentially explain its significance, map it to policies, and prepare a response for review.

The company’s open-source core adds another dimension. Open-source software lets technical buyers inspect code, host components, or contribute changes, depending on the project’s license and architecture.

Comp AI’s public positioning presents it as an alternative to Vanta and Drata. Enterprise coverage also identifies those companies as its most visible competitors.

The open-source approach can appeal to security teams that want greater transparency or deployment flexibility. It can also help developers evaluate integrations before making a wider commitment.

However, open code does not make a complete service transparent. Customers still need information about hosted models, data retention, prompts, access controls, and operational safeguards.

An agentic compliance platform also needs extensive integrations. It cannot assess current conditions if it sees only a fraction of a company’s systems.

Established vendors have invested heavily in those connections. Their relationships with auditors and consultants can also reduce friction during formal assessments.

Comp AI must therefore build two forms of trust. It needs technical trust from security teams and procedural trust from auditors, executives, and enterprise buyers.

Its reported customer growth suggests early demand, but customer count alone does not reveal deployment depth. A small audit-readiness account differs from a large continuous-security installation.

The 15-fold revenue growth figure also lacks a disclosed starting point. Rapid percentage growth is common when a company begins with a small revenue base.

That does not make the result unimportant. It means readers should treat it as a directional company metric instead of a complete measure of market position.

Comp AI’s expansion into security testing further complicates the comparison. Testing applications and infrastructure requires different expertise from collecting audit evidence.

Security testing must manage false positives, avoid disrupting systems, and distinguish theoretical vulnerabilities from exploitable risks. Customers will also expect clear records of every automated action.

Comp AI offers AI-powered penetration testing, according to Carhart. Penetration testing simulates attacks to identify weaknesses before hostile actors exploit them.

Automating portions of this work could widen access for smaller organizations. It could also increase testing frequency beyond the traditional annual engagement.

Still, autonomous testing introduces permission and containment concerns. A system must know which assets it may test, which methods are approved, and when to stop.

This is where the contest between execution and recordkeeping becomes concrete. A system of record can log a problem without changing production. An agent may interact with the affected environment.

Customers will judge that agent by more than task completion. They will examine whether its decisions are explainable, reversible, and properly authorized.

Incumbents can respond by adding similar agentic functions to existing platforms. They have customer data, established workflows, and resources to develop or acquire new capabilities.

Comp AI’s advantage is focus. Its architecture and product strategy begin with agents rather than adding them to older workflow assumptions.

Its disadvantage is the burden of proving that a newer system deserves broad access. The company must earn that trust while competing against vendors already embedded in procurement processes.

The $34 million round increases its capacity to compete. It also raises expectations that the company can move beyond audit acceleration without losing reliability.

What the Agentic Compliance Pitch Does Not Prove

Automating compliance tasks is useful, but autonomous security decisions demand stronger evidence, controls, and accountability.

The central uncertainty is not whether language models can draft policies. They can already produce plausible policy text from structured instructions and company documents.

The harder question is whether an agent can determine that a control works, especially when evidence is incomplete or contradictory.

A generated policy may look professional while misrepresenting actual practice. Automatically collected evidence may also be current but irrelevant to the exact control being tested.

Human reviewers often catch these mismatches through interviews, sampling, and professional judgment. An automated system must surface uncertainty instead of hiding it behind polished output.

Comp AI acknowledges the need for oversight. Carhart said humans review and approve drafted policies, with safeguards increasing as agents take more consequential actions.

That position is more credible than promising full autonomy immediately. It also reveals the main product tradeoff.

Requiring human approval reduces the risk of unsafe actions. Yet too many approval requests can recreate the workload that automation was supposed to remove.

A successful platform must place human review at meaningful decision points. It should not require approval for every routine evidence update or silently automate high-impact changes.

Permissions create another challenge. Continuous monitoring needs access to cloud services, identity systems, source code, ticketing tools, and business documents.

Those connections can make the platform valuable. They also make it a sensitive concentration point for security and privacy risk.

Customers need least-privilege controls, which limit each agent to the minimum access required. They also need tamper-resistant logs that show what an agent observed and changed.

Rollback procedures matter when actions modify live systems. The platform should distinguish recommendations from approved remediation and record who authorized each step.

Model behavior adds uncertainty. A language model can produce different answers from similar inputs, misunderstand organizational context, or assign excessive confidence to weak evidence.

Compliance outputs also require consistency. Auditors and customers may question a system that maps identical evidence differently across repeated assessments.

Continuous monitoring can produce alert fatigue if it treats every configuration change as urgent. Security teams already manage large volumes of findings from cloud and vulnerability tools.

Adding another stream of poorly prioritized warnings would weaken the value proposition. Comp AI needs to show that organizational context improves prioritization instead of adding narrative around existing alerts.

The company’s move toward security testing carries even greater operational risk. Automated tests must stay within approved boundaries and avoid exposing or damaging sensitive information.

These concerns do not invalidate the product direction. They define the requirements that separate a useful assistant from a trusted security operator.

Another unanswered question concerns independent validation. Comp AI reports customer growth and faster compliance outcomes, but publicly available coverage offers limited third-party performance data.

The claim that companies can reach compliance in days rather than quarters depends on their starting condition. A well-prepared software company differs from a complex enterprise with undocumented systems.

Certification timelines also involve auditors. A platform can accelerate evidence preparation, but it cannot guarantee the schedule or conclusion of an independent assessment.

Security buyers should ask for more specific measures. Useful evidence includes time saved per control, false-positive rates, remediation acceptance, and the percentage of actions requiring human intervention.

They should also examine how often automated findings survive auditor review. A high volume of generated evidence has little value if reviewers reject or repeatedly revise it.

Customer retention will provide another signal. Initial adoption can reflect curiosity or dissatisfaction with incumbents, while renewals show whether the workflow remains useful.

The Comp AI funding announcement does not include those measures. That is normal for a private financing disclosure, but it limits the conclusions outsiders can draw.

The company has demonstrated investor interest and reported fast adoption. It has not yet publicly demonstrated that agent-led continuous security works reliably across many complex environments.

Industry reporting places the round within a growing field of AI security and compliance startups. Security coverage also notes recent financing for companies working on agent certification, vulnerability management, and agentic GRC.

That crowded field can validate demand while increasing competition. It may also make customers cautious about granting sensitive access to young vendors.

Comp AI’s challenge is to prove that its agents reduce work without weakening review. If it succeeds, compliance platforms will become more active participants in security operations.

If it fails, buyers may retain agentic features for drafting and triage while keeping validation and remediation under direct human control.

Three Signals That Will Define Comp AI’s Next Stage

The next test is whether Comp AI can convert funding, customer claims, and product ambition into measurable security outcomes.

The first signal is the delivery of continuous control validation across real customer environments. Comp AI has described this capability as a major destination for the new capital.

Buyers should watch for specific releases rather than broad agentic language. Useful details include supported systems, monitored control types, approval boundaries, and action logs.

A credible release would show how the platform identifies a change, connects it to a control, validates the impact, and routes the response.

It should also explain when the system stops and requests human judgment. Clear boundaries would strengthen the company’s claim that agents can perform work safely.

A vague product expansion would weaken that claim. Rebranding existing evidence collection as continuous security would not establish a meaningful technical difference.

The second signal is independent acceptance by auditors and mature security teams. Comp AI’s agents can produce evidence, but external reviewers decide whether that evidence supports an assessment.

Published customer cases should report concrete outcomes. These might include fewer manual evidence requests, faster remediation, or higher acceptance of automatically assembled materials.

Regional reporting says Comp AI is moving toward larger organizations with more complex security needs. That move will test the product beyond startup audit preparation.

Large organizations have fragmented ownership, legacy systems, multiple cloud environments, and formal change controls. They also require detailed vendor risk reviews.

Successful deployments there would strengthen the case for Comp AI’s organizational-context approach. Persistent reliance on high-touch services would suggest the agents still need extensive human support.

The third signal is the competitive response from Vanta, Drata, and adjacent security platforms. These vendors do not need to copy every feature to narrow Comp AI’s opening.

They can add agents to existing evidence workflows, expand automated remediation, or deepen partnerships with security-testing providers.

Their installed customer bases provide distribution advantages. Comp AI must release useful capabilities quickly without weakening governance.

Competitive reactions will also reveal whether continuous agentic compliance is becoming a standard product category. If incumbents emphasize similar execution features, Comp AI’s thesis gains validation.

However, category validation does not guarantee startup leadership. Features can converge quickly when they depend on widely available models and common enterprise integrations.

Comp AI needs differentiation that extends beyond model access. Its open-source core, workflow design, organizational context, and customer experience are potential sources of durability.

Security outcomes will remain the decisive measure. Buyers will care whether the product identifies meaningful risk, reduces repetitive work, and preserves accountability.

The financing gives Comp AI resources to pursue that standard. Its reported growth gives the company a customer base for testing and refinement.

Neither factor removes the central tradeoff. Security teams want faster, continuous validation, but they cannot delegate responsibility to opaque automation.

That makes Comp AI’s expansion a useful test for the wider agentic software market. Many vendors promise agents that do work rather than merely answer questions.

Compliance offers a demanding proving ground because every action needs evidence, permissions, and review. Errors can affect audits, contracts, sensitive data, and production systems.

The best outcome is not maximum autonomy. It is carefully bounded automation that makes control failures visible sooner and routes decisions to the right people.

Readers evaluating the Comp AI funding story should therefore follow product evidence, not the financing headline alone. Watch how often agents act, how often humans intervene, and whether auditors accept the results.

If Comp AI publishes credible answers, its move from compliance automation into continuous security will look justified. If those answers remain vague, the company may remain a faster audit-preparation platform with a larger ambition.

The next few months should clarify which version is emerging. Security leaders should ask vendors for documented boundaries, independent validation, and real operating metrics before expanding agent permissions.