Glow

**Product and the concrete problem it solves.** Glow attacks a visibility-and-control gap that widened sharply as generative AI moved onto employee laptops. The endpoint is where an organization's software actually executes, and it is now where employees unilaterally introduce AI coding assistants, autonomous agents, browser extensions, MCP connectors, and open-source packages — often within hours of a tool going viral, and always ahead of procurement. Glow's own framing of the shift is that the share of corporate devices running AI software rose from roughly 15% to roughly 45% within a year, and that a large majority of what runs on a managed device (the company claims 67%) is invisible to the tools already installed on it. The consequence is that the endpoint agent stack an enterprise pays for — EDR, MDM, DLP — reports on a fraction of reality, and its central assumption, that the security team will detect and respond after something malicious runs, breaks when the risky artifact is a legitimate-looking npm package, an unvetted AI agent with broad local permissions, or a developer tool with credential access. Glow's product is therefore built around three functions the company names publicly: **Asset Intelligence** (continuous discovery and validation of endpoints and every software layer on them, maintaining what it calls a living inventory that stays accurate), **Software Control** (autonomous remediation and policy enforcement agents that decide what is allowed to enter and what gets removed), and **Safe AI Adoption** (shadow-AI discovery plus governed enablement of sanctioned AI tooling). The posture is prevention rather than post-hoc detection: keep the risky artifact out, rather than alert on it after execution.

**Core technology and how it actually works.** Glow's architecture pairs a device-resident sensing and enforcement layer with a cloud reasoning layer built from specialized AI agents rather than a single monolithic model. Those agents continuously map the environment, score risk in real time against enterprise context, and then act — quarantining, removing, or gating software without a human writing a rule per artifact. CEO Roi Tiger has publicly drawn the contrast with incumbent EDR: existing products focus primarily on detecting threats after they emerge, whereas Glow is designed to prevent risky software, AI agents, and developer tools from entering enterprise environments in the first place. On the model layer, Glow has said it uses frontier models from Anthropic and Google (Gemini) served via Amazon Bedrock, combined with proprietary software that supplies the enterprise-specific context — an important diligence detail, because it locates the company's defensible work in the context/reasoning engine, the device telemetry, and the enforcement plumbing rather than in model training. Publicly cited early results are operational rather than benchmark-style: the platform has blocked malicious npm packages before installation, surfaced rogue AI agents running on employee machines, and identified devices whose existing endpoint tooling was missing or silently degraded — the last of which is a meaningfully differentiated capability, since a decayed EDR agent is a failure mode most EDR consoles report poorly on themselves. Typical deployments are described as spanning tens of thousands of devices. The company lists SOC 2, ISO 27001, and GDPR compliance on its site.

**Market, customers, and go-to-market.** Glow is selling into the endpoint security market, which press coverage of the launch sizes at roughly $40 billion, against incumbents with entrenched agents and multi-year enterprise contracts. Its wedge is not to rip out EDR immediately but to occupy the adjacent, newly urgent budget line — AI governance and software control on the endpoint — where CISOs face board-level pressure and no incumbent has a settled answer. Go-to-market is a classic Cyberstarts-style, CISO-led enterprise motion, and the company's public proof points reflect it: named customer logos on its site include Fanatics, Qualtrics, Antares Capital, Xactly, and Hospital Sisters Health System, with attributed endorsements from Larry Dolan (SVP, CISO at Fanatics), Kyle Weckman (CISO, Antares Capital), Eli Edelkind (CISO, CAVA), and Scott Crowder (SVP & CIO, BMC Software). Sector concentration so far is healthcare, retail, and financial services — regulated, device-heavy industries where unmanaged software is both a compliance and a breach problem. COO Emily Heath, herself a former CISO at United Airlines and DocuSign, is a strong signal of a buyer-empathy-led motion. Glow showcased the platform publicly at Black Hat USA 2026 (3-5 August, Las Vegas), roughly two weeks after emerging from stealth.

**Traction, funding, and third-party validation.** Glow disclosed $180 million at a $1.2 billion valuation on 22 July 2026, led by Sequoia Capital, Cyberstarts, Greenoaks, and Redpoint Ventures, with participation from Index Ventures, Swish Ventures, Lux Capital, Operator Collective, and Holly Ventures. Reporting differs on the round structure and this is worth flagging precisely: the company's own announcement and several outlets present the $180M as a single headline raise (some describing it as a Series A), while CTech reports it as three stacked rounds disclosed together — a roughly $20M seed from Sequoia and Cyberstarts, a roughly $60M Series A at about a $400M valuation with Index and Greenoaks, and a roughly $100M Series B at the $1.2B mark. Either way the substance is the same: a company founded in 2025 was marked up roughly 3x within about a year and never publicly shipped a product until launch day. Headcount is close to 100, with roughly 65-70 people in Israel and the remainder in the United States. The strongest external validation is the combination of the syndicate (Sequoia plus Cyberstarts, whose CISO network is the sharpest early-signal machine in Israeli cyber), the named enterprise logos with on-record CISO quotes, and the seniority of the operating team. What is not disclosed is revenue, ARR growth, customer count, or net retention — the standard gaps for a company at this stage of publicity.

**Founders and team background.** The founding team is unusually senior for a one-year-old company. CEO and co-founder **Roi Tiger** spent nine years at Meta, most recently as VP of Engineering, arriving through Meta's acquisition of Onavo, the Israeli mobile-analytics company he co-founded — a background that is directly on-point (endpoint telemetry at consumer scale) and also a reputational liability, since Onavo's data-collection practices became a well-documented controversy for Meta. CTO and co-founder **Omer Singer** was Snowflake's Head of Cybersecurity Strategy and a Unit 8200 veteran, giving the company depth in security data modeling and in selling to security-data buyers. VP of R&D and co-founder **Ophir Arie** was VP of R&D at Claroty, a Talpiot and Unit 8200 graduate, bringing hard-edged experience building agents and asset-discovery systems for environments where you cannot simply reboot the endpoint. The executive bench extends the pattern: CPO **Arnon Joseph** spent eight years at Meta as a Senior Director of Product, and COO **Emily Heath** was CISO at United Airlines and DocuSign, a Cyberstarts partner, and a Wiz board member through the $32 billion Google acquisition. This is a team assembled to sell to Fortune 500 CISOs from day one.

**Competitive dynamics.** Glow is entering the most heavily defended segment in security software, and its edge is timing and architecture rather than a novel primitive. (1) Against **CrowdStrike** and **SentinelOne**, the incumbent EDR/XDR duopoly, Glow argues category mismatch — detection-and-response versus prevention-and-control — but both incumbents have enormous distribution, their own agent already installed, and every incentive to ship AI-governance features into an existing console. (2) Against **Microsoft Defender for Endpoint** and Intune, the bundling threat is structural: 'good enough and already paid for' has killed better point products before. (3) Against **Tanium** and **Nexthink**, Glow's Asset Intelligence pillar competes directly on endpoint discovery and software inventory, a mature market where the incumbents' claim is completeness at scale. (4) Against the shadow-AI cohort — enterprise-browser and AI-governance players such as **Island**, **Prompt Security**, and a rapidly growing set of AI-agent security startups including Israeli peers **Koi Security**, **Neo**, and **Bloom Security** — Glow competes for the same emerging budget line with a device-layer rather than browser-layer or identity-layer approach. (5) The quiet fifth competitor is **the endpoint agent budget itself**: security teams resist adding another agent, so Glow must either displace something or prove it consolidates. Its most defensible differentiators are the control-plane position (allow/deny at install time), the ability to audit the health of other endpoint agents, and a context engine tuned to AI-era artifacts that legacy application-control products (which think in executables, not agents and MCP servers) do not model well.

**Defense, security, and resilience dual-use relevance.** The dual-use case here is real but should be stated as infrastructure relevance rather than a fielded defense capability. Software-supply-chain compromise and unvetted autonomous agents on privileged endpoints are among the most consequential vectors facing government, defense-industrial, and critical-infrastructure operators — the class of problem exemplified by malicious package injection and by AI agents holding credentials with lateral reach. A prevention-first control plane that maintains an accurate inventory of everything executing on a device, blocks unvetted packages at install time, and detects silently degraded security agents maps directly onto the hardening requirements of defense primes, government IT estates, hospital systems, utilities, and other operators of essential services; the healthcare and financial customers already disclosed are the commercial expression of the same requirement. The team's provenance reinforces the adjacency: Unit 8200 and Talpiot backgrounds, and a co-founder from Claroty, whose entire business is protecting industrial and critical-infrastructure environments. Calibration matters, though. As of August 2026 there is no disclosed government, defense, or critical-infrastructure contract, no FedRAMP or IL-level authorization in the public record, and no Israeli MoD or allied-military engagement announced. Glow is dual-use in the same sense that endpoint security generally is — the same product hardens a bank and a defense agency — not because it has built anything military-specific. Any national-security thesis rests on the sovereignty and resilience value of allied-controlled endpoint control planes and on future public-sector accreditation, both of which remain to be demonstrated.

**Growth stage, trajectory, and key diligence risks.** Glow is an early-stage company with late-stage resources: founded 2025, ~100 employees, a shipped and deployed product, marquee logos, and $180M raised at a $1.2B valuation before its first Black Hat. The bull case is straightforward — an exceptional team, an unambiguous and newly urgent problem, a control-plane position with natural expansion room, and the best-networked syndicate in Israeli cyber. The diligence risks are equally clear. (1) **Valuation-ahead-of-evidence risk**: a $1.2B mark on a company with no disclosed revenue is a bet on the team and the category, and it sets a high bar for the next round. (2) **Incumbent bundling risk**: Microsoft, CrowdStrike, and SentinelOne can ship overlapping AI-governance capability into agents that are already deployed, compressing Glow's window. (3) **Agent-fatigue and displacement risk**: enterprises resist a fifth endpoint agent; Glow must consolidate spend, not add to it, and consolidation sales are slower. (4) **Model-dependency and cost risk**: reliance on third-party frontier models via Bedrock exposes unit economics and behavior to vendors Glow does not control, and puts the durable moat in context and enforcement rather than in the model. (5) **Prevention false-positive risk**: an autonomous system that removes or blocks software will eventually block something a business needs; a single high-profile productivity incident is a serious enterprise-sales setback, and the 2024 CrowdStrike outage set the industry's tolerance for endpoint-agent failure very low. (6) **Founder reputational risk**: the Onavo history will surface in procurement conversations at privacy-sensitive and public-sector buyers, precisely the accounts a dual-use thesis depends on. (7) **Disclosure gaps**: revenue, customer count, retention, and the exact round structure are not confirmable from public sources. Progression toward 'mid' would be evidenced by disclosed ARR or customer counts, a public-sector or defense accreditation, displacement wins against an incumbent EDR, and evidence that the prevention model holds up operationally at six-figure device counts.