Obsidian Security

1.1 Identity and business model

Mandos and The Next Web both place Obsidian's founding in 2017, while Mandos and the company's August 2026 release identify Palo Alto, California as its headquarters. The company's current positioning is narrower than a generic cloud-security label: it says it governs data, AI, and agents operating inside third-party applications, and its August release describes runtime controls intended to detect or block excessive access and policy violations. That describes the product thesis, not independently demonstrated prevention efficacy. The assigned evidence supports a private-company classification and a latest reported Series D, but does not establish a complete product portfolio, audited deployment outcomes, or a current geographic footprint.[CO001, CO002, CO003]

1.2 Leadership and governance

Mandos lists Hasan Imam as CEO on a profile updated in October 2026, and SiliconANGLE identifies him as chief executive in its August funding coverage. Menlo Ventures' investor perspective names founders Matt Wolff, Glenn Chisholm, and Ben Johnson and records historical operating roles: Wolff as chief scientist, Chisholm as chairman and CPO, and Johnson as CTO. Menlo is an investor, so that account is useful relationship evidence rather than an independent current board roster. The currently assigned pool does not establish whether those historical roles remain unchanged, the full executive team, board composition, voting control, or succession coverage. Those limits matter because the company remains private and the Series D investor list does not disclose ownership or governance rights.[CO004, CO005, CO006]

1.3 Financing, valuation, and operating scale

Obsidian announced an $85 million Series D led by Crescent Cove Advisors on August 4, 2026; SiliconANGLE independently reported the same round and $1.1 billion post-money valuation. Mandos' October profile reports $204.5 million of priced equity across four rounds, while Caplight's preview separately lists an unexplained “Early Funding” entry; the total and round count bases are not fully reconciled. The company says more than 100 customers spend over $100,000 annually, more than 14 spend over $1 million, and 60 Fortune 500 companies are customers. These are company-reported commercial indicators, not audited revenue, ARR, retention, or customer concentration. The release and Caplight's preview name overlapping backers, but neither provides ownership shares or control rights. Mandos estimates approximately 269 employees as of September 2026; no source supplies an audited headcount, burn, runway, or operating margin.[CO007, CO008, CO009, CO011, CO012, CO013, CO014]

1.4 Milestones and unresolved diligence

The funding record shows a progression from early priced rounds to the unicorn-marked Series D, alongside a product shift toward agent governance. Caplight's public preview supplies dates for Series A through Series D and an “Early Funding” entry, but displays no amounts and does not explain that 2025 entry. Obsidian announced SaaS AI agent defense in September 2025 and said in August 2026 that it was extending native governance controls to Claude Code and Cowork; these are product-announcement milestones, not proof of customer adoption or control effectiveness. The Next Web adds a valuation caution, noting that not every AI-security startup will grow into its price. No assigned source establishes a company-specific lawsuit, regulatory action, outage, or security incident; that is an evidence limitation, not evidence that none exists. The dated timeline is therefore selective rather than a complete corporate history.[CO001, CO007, CO010, CO015, CO016]

2.1 Market boundary and substitutes

Obsidian operates across overlapping categories rather than one cleanly measured market. The narrowest useful boundary is software that discovers and governs identities, permissions, configurations, and agent activity inside third-party SaaS applications; broader non-human-identity (NHI) estimates include application, workload, device, and cryptographic identities, while agent-security estimates can include model, runtime, and governance tools beyond SaaS. These adjacent categories are useful demand signals, not additive addressable spend. The fetched SSPM market summary describes licensed or subscription software for SaaS discovery, configuration, identity/access risk, compliance, and remediation, but it also flags overlap with IAM, CASB, and CSPM. Manual audits, native application controls, identity platforms, and broader security suites are therefore plausible status-quo substitutes; the evidence does not quantify their displacement or incremental budget.[CM001, CM004, CM005, CM009]

2.2 Sizing lenses and limits

Available published estimates vary by category, period, and method. MarketsandMarkets reports a global agentic-AI-security estimate of USD 1.65 billion for 2026 and USD 13.52 billion for 2032, with a 42.0% CAGR over 2026-2032; Mordor's broader NHI series is USD 8.22 billion for 2026 and USD 22.94 billion for 2031. The GII-hosted SSPM summary reports USD 484.4 million for 2025 and USD 3.53 billion for 2030, while SNS Insider gives inconsistent 2025 AI-agent-security figures on the same page. These are third-party forecasts, not audited spending, and their scopes are not reconciled. They establish category interest, not the amount Obsidian can capture. No supported public data isolates its serviceable market, pricing, attach rate, or obtainable share; the figure is a conceptual scope map, not a proportional market-size pyramid.[CM006, CM007, CM008, CM009, CM010, CM011, CM012, CM013]

2.3 Buyer segments, adoption drivers, and constraints

The clearest buyer hypothesis is a large enterprise with a sprawling SaaS estate and security teams accountable for identity, access, and compliance risk. A profile attributes 60 Fortune 500 customers to company materials, and a separate syndicated announcement describes Fortune 500 and Global 2000 organizations as a target; these statements show enterprise orientation, not independently measured category penetration. Dealroom reports the CEO saying nearly 70% of Obsidian customers already let agents interact with business data, a company-level signal rather than an industry adoption rate. Analyst summaries identify application sprawl, unmanaged SaaS, identity exposure, and regulatory pressure as adoption drivers, while also citing integration complexity, skills shortages, budget constraints, and overlap with adjacent tools. Actual payer, procurement path, willingness to pay, and buying trigger by segment remain unverified; the buyer map distinguishes reported segmentation from inferred budget ownership.[CM014, CM015, CM016, CM017, CM019]

3.1 Competitive field and evidence boundaries

The assigned evidence points to three competitive layers: SaaS-security specialists, agent-security specialists, and broader security platforms that could extend or bundle into the same buying decision. A third-party profile names Grip Security, AppOmni, Adaptive Shield, CrowdStrike, Microsoft, Okta, Vorlon, and Valence Security as competitors, and reports that an analyst rates Obsidian alongside several of those vendors as a leader. Caplight's embedding-similarity list separately surfaces AppOmni, Reco, and Delinea; that is a similarity signal, not a feature benchmark. Menlo Ventures' 2022 investor post names SaaS-native controls, CASBs, and SIEMs as then-current partial substitutes; that is historical context, not proof of present-day buyer behavior. The sources provide the most detailed product comparison for Obsidian and Zenity. For other named alternatives, product scope, price, scale, and customer overlap remain unverified in this pool. Internal build is a plausible buyer path, but the assigned sources do not measure it.[CP001, CP002, CP011, CP012, CP013]

3.2 Specialist capabilities and strategic adjacency

Obsidian's reported differentiation combines security for third-party SaaS applications with controls over agent actions in those applications. SiliconANGLE describes runtime enforcement against privilege escalation, excessive data access, and policy violations, while reporting coverage for agent-building and developer environments. Calcalist describes Zenity evaluating an agent action's intent before execution and allowing a security team to approve, modify, or block it; it also lists support for several commercial and developer-agent environments. Both therefore make a pre-execution control claim, but the sources do not offer a head-to-head efficacy test or a common definition of coverage. The SNS Insider also reports AI-agent security additions by Microsoft, Google Cloud, Palo Alto Networks, and CrowdStrike; its market-report descriptions are not independent product tests. The analyst rating reported by CyberCompanyProfiles and the separate similarity list are category signals, not evidence that products perform equally or are direct substitutes in each buyer workflow.[CP003, CP009, CP010, CP002]

3.3 Pricing, lock-in, and displacement risk

Public scale signals are not directly comparable across the two specialists. SiliconANGLE reports Obsidian's August 2026 Series D at an $1.1 billion post-money valuation and more than 100 customers spending over $100,000 annually, with more than 14 above $1 million; these are company-reported customer-spend thresholds, not disclosed revenue or ARR. Calcalist reports Zenity's $125 million Series C and approximately $185 million in total funding, while describing its customer base as large enterprises. CyberCompanyProfiles reports Obsidian's free tier and quote-based paid tiers, but the assigned evidence lacks comparable competitor prices, discounts, contract terms, churn, multi-homing, or migration costs. Its warning that large platforms could bundle alternatives is a plausible displacement risk, not a measured outcome. Buyer references, renewal data, and controlled win/loss evidence are needed before asserting a durable moat.[CP004, CP005, CP006, CP007, CP008, CP014, CP015]

4.1 Commercial traction is visible, but not revenue

SiliconANGLE reported that CEO Hasan Imam described more than 100 customers paying over $100,000 per year and more than 14 paying over $1 million annually; Unite.AI and other coverage repeated those company-reported tiers. These are customer-spend thresholds, not an aggregate revenue figure, ARR, recognized revenue, or a customer count distribution. Unite.AI separately reported that 60 Fortune 500 companies were in the customer base, again without a disclosed denominator for deployment scope, contract value, or timing. The evidence therefore supports enterprise reach and a high-spend cohort, but not revenue concentration, growth rate, renewal quality, or realized pricing. The available materials also do not describe distinct revenue streams, contract terms, or accounting treatment. The table keeps those evidence-backed customer tiers separate from the private metrics needed to convert them into an underwriting-grade revenue bridge.[CI003, CI004, CI005, CI014]

4.2 Recurring economics and sales efficiency remain unpriced

Neither the disclosed customer-spend bands nor the financing coverage establishes gross margin, net revenue retention, churn, customer concentration, customer acquisition cost, payback, sales-cycle length, or service-delivery cost. Menlo Ventures' 2022 portfolio-investment post said Obsidian had seen 5x growth in deals of $100,000 or more over the preceding year. That is a historical deal-count growth proxy from an investor with a portfolio relationship; it is not current bookings growth, revenue growth, CAC efficiency, or payback. No pricing schedule or realized discount data in the assigned materials lets a reader translate annual customer spend into list price or recognized revenue. As a result, the market-facing evidence is stronger on customer scale than on revenue quality or margin path. The unit-economics table marks each unavailable field as a diligence request rather than filling gaps with SaaS assumptions.[CI009, CI014, CI015]

4.3 Capital raised does not establish runway or valuation support

SiliconANGLE reported an $85 million Series D led by Crescent Cove Advisors at a $1.1 billion post-money valuation on August 4, 2026; Unite.AI reported total funding above $200 million after the round. The proceeds were described as supporting research and development and go-to-market investment, while Dealroom reported that the company said the funding should carry it to positive cash flow. No source in this assigned evidence quantifies current cash, burn, debt, working capital, capex, or the timing and assumptions behind that target. Funding raised is a historical flow, not cash currently available, and the target is not evidence that positive cash flow has been reached. The Next Web's caution that not every AI-security company will grow into its price is relevant downside context; without company revenue or margins, no revenue-multiple or cash-runway conclusion is supportable.[CI001, CI002, CI006, CI007, CI008, CI013, CI015]

4.4 Exhibits

5.1 Product surface and customer workflow

Obsidian describes a security layer for third-party business applications that now also governs AI agents acting in those environments (CE001). Its August 2026 announcement names controls for Claude Code and Cowork, including permission management, restricted access to sensitive systems, and monitoring of interactions in business applications (CE002). The same announcement says runtime protection can detect and block excessive data access, privilege escalation, and policy violations during task execution, and that the product inventories MCP servers and large language models (CE003, CE004). These are company-described features, not independently measured outcomes. The January SaaS-supply-chain launch broadens the described workflow from individual account and configuration monitoring to SaaS-to-SaaS integration risk: visibility, prevention, early detection, and incident containment (CE005). SiliconANGLE reports identity and activity normalization across SaaS applications and a knowledge graph of usage patterns (CE006); The Next Web describes monitoring for compromised accounts, risky configurations, and identity-based attacks (CE007). The module matrix and workflow table distinguish these advertised functions from unknown production maturity and unreported customer outcomes.[CE001, CE002, CE003, CE004, CE005, CE006, CE007]

5.2 Operating model and technical boundaries

Public descriptions suggest an operating model that gathers SaaS identity and activity context, normalizes it across applications, and uses behavioral analysis to surface deviations (CE006, CE008). Menlo Ventures described that approach in an investment article; its investor relationship and the article's undated, historical framing make it useful as product-history context, not independent current validation (CE008). The company’s newer materials position the same SaaS context as a control point for agents and integrations (CE001, CE005). This is a description of product intent, not a verified architecture diagram: the assigned sources do not specify connector coverage, ingestion or retention design, enforcement points, deployment model, or service-level reliability (CE012). MCP broadens the trust boundary beyond an agent to its host, client, server, tools, and data. OWASP identifies prompt/tool poisoning, changing tool definitions, confused-deputy behavior, over-scoped permissions, and supply-chain risks; the Cloud Security Alliance's March 2026 guide is explicitly a draft and says conventional application-security controls are insufficient for the protocol's attack surfaces (CE009, CE010). These sources define diligence tests for Obsidian's MCP inventory and runtime controls; they do not establish that the company has or lacks any specific control.[CE001, CE005, CE006, CE008, CE009, CE010, CE012]

5.3 Maturity, differentiation, and proof gaps

The allocated evidence supports a coherent adjacency: a product already described as monitoring SaaS identities and activity is being extended to agent permissions, runtime policy, and integration risk in the same application estate (CE001, CE005, CE007). UpGuard, a competing security vendor, also characterizes Obsidian as combining SSPM, integration risk, AI governance, and identity-threat detection; its comparison is vendor-authored and is not an independent benchmark (CE011). The sources do not show comparative detection quality, verified prevention rates, false-positive rates, coverage by application or agent, or customer-validated time savings (CE012). Deployment and operational maturity remain open: the public descriptions in this assigned pool do not establish integration counts, implementation time, uptime, incident-response commitments, certification scope, or privacy and retention controls (CE013). The pool also contains no company repository activity, package downloads, or practitioner-community measurements, so developer adoption cannot be assessed (CE014). Treat product breadth and the claimed centralized control plane as a thesis to test with architecture documentation, a live technical evaluation, and customer references, not as evidence of a durable technical moat.[CE001, CE005, CE007, CE011, CE012, CE013, CE014]

5.4 Diligence priorities

The highest-value technical diligence is to test agent and integration controls against realistic permission escalation, excessive data access, tool-definition changes, prompt injection, and cross-application movement (CE003, CE009, CE010). Request application-by-application connector and action coverage; evidence of enforcement points and auditability; test methodology and false-positive/negative rates; and production references that can distinguish a shipped control from a roadmap or launch claim (CE002, CE003, CE012, CE013). For MCP, verify discovery completeness, change detection, authorization scope, and whether policy is enforced outside the agent's own tool descriptions. These requests address the actual dependency chain rather than assuming that inventory alone provides prevention. No public developer-signal evidence was available in the assigned pool. That is a research limitation, not proof that the product has no developer ecosystem or that no security incident has occurred (CE014). Ask for release cadence, documented integration/version support, independent penetration or control testing, and incident history under appropriate confidentiality. Until those records and customer-side validation are available, product breadth is more supportable than claims about efficacy, reliability, or defensibility.[CE002, CE003, CE009, CE010, CE012, CE013, CE014]

5.5 Exhibits

6.1 Customer segments and disclosed scale

The available customer evidence is weighted toward large enterprises rather than a disclosed count of all active users or deployments. Sacra estimates more than 100 customers spend over $100,000 annually and more than 14 spend over $1 million; SiliconANGLE separately attributes the same thresholds to CEO Hasan Imam, while FinTech Global repeats the spending bands in coverage of the Series D. These are useful indicators of paid-account scale, but they are not audited customer counts, do not disclose the period-end measurement date, and do not show contract renewal. Sacra and a company-sourced Unite.AI article also report 60 Fortune 500 customers. None of the available sources names those accounts or maps the spend tiers to deployment scope, so the evidence supports enterprise reach more strongly than customer-level production proof or concentration analysis.[CU001, CU002, CU003, CU016]

6.2 Use cases, deployment evidence, and outcomes

The clearest first-person usage evidence is a PeerSpot reviewer describing incident response work across client environments, especially Microsoft 365 business-email compromise investigations. The reviewer says Obsidian helped surface anomalous accounts and hardening actions, and reports that some investigations that might previously have taken days were reduced to a day or less; this is a single customer's account, not a measured average or independently validated outcome. Sacra describes three anonymized customer examples: a Global 2000 telecom account identifying about $1.95 million in redundant-license savings, another customer reporting an 85% reduction in manual SaaS-security work across more than 100 tenants, and a hospitality customer reporting an initial picture of a security event within five minutes across more than 300 cloud applications. The sources do not identify these customers, define measurement methods, or provide before/after validation; outcomes should therefore be treated as reported case evidence rather than generalized ROI.[CU007, CU008, CU009, CU013, CU014, CU016]

6.3 Adoption, expansion, and durability limits

Company-linked coverage frames agent access as an expansion driver: FinTech Global quotes Imam saying more than 70% of customers already let agents into third-party applications, Dealroom reports his estimate as nearly 70%, and SiliconANGLE separately reports 65% of enterprise customers. Those figures have different stated denominators and are not reconciled, so they should not be treated as one consistent adoption rate. Menlo Ventures reported 5x growth in deals of $100,000 or more during the year preceding its April 2022 investment post, but that is historical deal growth, not a current customer cohort or retention metric. Public review evidence is directionally positive but bounded: Gartner Peer Insights displays a 4.8 rating across 28 ratings, while a surfaced review and the PeerSpot account include useful product feedback as well as integration and interface improvement requests. No assigned source establishes NRR, GRR, churn, renewal, revenue concentration, or the conversion of pilots into production deployments.[CU004, CU005, CU006, CU010, CU011, CU012, CU015]

7.1 Runtime controls face an evidence gap

The assigned evidence supports a high-impact threat model, but not a company-specific security failure. Token Security describes a separate Salesloft-Drift OAuth compromise that reached Salesforce customer environments; The Next Web describes how an over-permissioned or hijacked agent could move quickly across enterprise applications. Neither source reports an Obsidian incident, and Token Security is a competing vendor promoting its own controls (CR004, CR005). Obsidian's runtime enforcement is described in company-issued material and a CEO interview, while the WorkOS comparison says the product monitors activity after authentication. WorkOS is a competitor, not an independent benchmark; the pool contains no deployment test, independent red-team result, false-positive rate, or enforcement-latency measurement (CR001, CR002, CR003). Treat prevention claims as unverified until customer evidence demonstrates coverage, fail-open behavior, and response under machine-speed activity. The matrix therefore separates unknown likelihood from potential impact rather than assigning incident probabilities (FR001). The operational register is deliberately ranked by potential consequence and identifies the evidence needed to assess mitigation maturity (TR002).[CR001, CR002, CR003, CR004, CR005]

7.2 Legal and regulatory posture remains unverified

The assigned pool contains no regulator, court, certification, contract, or company compliance record that establishes Obsidian's current legal posture. This is a scoped evidence gap, not evidence that litigation, enforcement, or a compliance failure exists. Privacy and AI-governance obligations will depend on the data processed, customer use cases, and the company's contractual role; those facts are not established in this source set (CR012). The regulatory register is a partial list of diligence scenarios, not a finding that any specific regime applies. Until current counsel-reviewed applicability and control evidence are supplied, both likelihood and mitigation maturity remain unassessed and residual exposure is material (TR001). Obtain current legal and compliance records before relying on certifications, customer representations, or applicability conclusions.[CR012]

7.3 Customer, platform, and financing dependencies need diligence

Reported customer and financing indicators do not resolve concentration or cash-availability risk. SiliconANGLE attributes to CEO Hasan Imam claims that more than 14 customers paid over $1 million annually and more than 100 paid over $100,000 per year as of the August 2026 interview; the report gives neither a total customer denominator nor customer revenue shares, and the tiers may overlap (CR014). SiliconANGLE and The Next Web both report an $85 million Series D at a $1.1 billion post-money valuation, but neither metric is cash available or runway (CR006, CR007). Dealroom says the company expects the proceeds to carry it to positive cash flow; that is a reported expectation, not a cash-flow result (CR011). Meanwhile, a market report lists major platform vendors expanding agent-security capabilities, and SiliconANGLE reports a substantial adjacent-market round, increasing execution and competitive pressure (CR009, CR013). The dependency register separates customer, SaaS-platform, and capital-provider exposures (TR003); diligence should obtain concentration, renewal, integration, burn, and runway schedules.[CR006, CR007, CR009, CR011, CR013, CR014]

8.1 Entry conclusion and price discipline

Treat the reported $1.1 billion post-money Series D mark as transaction context, not as a verified enterprise-value multiple or a public-market clearing price. SiliconANGLE, Dealroom, and The Next Web report an $85 million round; Unite.AI reports more than $200 million raised across five rounds, which is capital raised rather than cash available. None of those figures supplies the current ARR, recognized revenue, growth, gross margin, retention, burn, runway, or security-level financing terms needed to calculate an Obsidian-specific multiple or preference-adjusted return. The company site also presents product and ROI claims, but those do not substitute for audited recurring economics. I recommend research-more before making a price-based investment decision: confidence in the price assessment is low, and valuation underwriting risk is high because the transaction mark cannot be tested against its financial denominator or common-equity rights. A verified ARR bridge, cohort metrics, cap table, and Series D documents could change the call; until then, neither a buy case nor a numeric downside range is supported.[CV002, CV012, CV013, CV014, CV015]

8.2 Investment thesis and anti-thesis

The product thesis is coherent: Obsidian's site describes real-time visibility, continuous governance, and runtime enforcement across agents, identities, and integrations in third-party apps, while The Next Web describes the existing SaaS layer as a potential advantage in governing agents there. This is evidence of positioning, not independent proof of efficacy. The financing coverage also relays large-customer signals: Unite.AI attributes more than 100 customers spending over $100,000 annually and more than 14 spending above $1 million to the company. SiliconANGLE and Unite.AI report different agent-access percentages and denominator wording, so those figures should not be combined into one adoption rate. Menlo Ventures' investor-authored 5x deal-growth statement is historical and interested-party evidence, not current ARR or retention. The anti-thesis is therefore about underwriting rather than product relevance: spend thresholds, a funding headline, and seller or investor positioning do not establish recurring revenue quality, margins, customer concentration, or returns at the mark.[CV001, CV003, CV004, CV005, CV006, CV010, CV011]

8.3 Scenarios, comparables, and diligence gates

The available comparison set is directional rather than a valuation formula. Windsor Drake's April 2026 report gives a public cybersecurity median on an NTM-revenue basis and a much wider revenue range for AI-native private security, but it is the advisory firm's own synthesis and does not show that Obsidian meets the relevant growth, margin, or architecture conditions. TechCrunch's reported Cyera financing multiple is competitor-specific, based on unnamed sources, and accompanied by a company denial; it is not an independent benchmark for Obsidian. The Next Web separately cautions that security valuations have run hot. The bull, base, and bear cases below therefore state conditional assumptions and downside triggers, but leave numeric equity ranges and probabilities unassigned rather than inventing them. Diligence should obtain current ARR and revenue reconciliation, growth, gross margin, retention and concentration; cash burn, runway and debt; fully diluted capitalization and preference terms; and customer evidence for agent-product adoption. Revisit entry only after normalizing those inputs against genuinely comparable transactions and modeling dilution, preferences, and exit proceeds.[CV007, CV008, CV009, CV012, CV013, CV014, CV015, CV016]