Reco's $55M round crystallizes AI-agent security as its own venture category — and the PANW/ZS/CRWD bundle math gets harder

Reco's pitch is anchored in a number that, until2026, didn't appear in any vendor's marketing: at a Fortune 100 customer, Reco's discovery graph found 21,000 AI agents the company didn't know about. At a separate large financial-services customer, it surfaced an agent an ex-employee had stood up that could still reach Salesforce and exfiltrate data to an external domain. Reco says 40% of the agents it discovers have a toxic combination — sensitive file access plus internet exposure — and that enterprises routinely run hundreds of thousands of agents once you multiply headcount by the \"a hundred agents per person\" baseline many corporate AI strategies imply.

Strategic investors writing checks is the most reliable leading indicator of which problem is actually keeping a Fortune500 CISO up at night. AT&T's venture arm led the round and AT&T is a paying customer. The reasoning, spelled out by AT&T Ventures in the announcement and corroborated by SecurityWeek and GovInfoSecurity, is that enterprises must understand and control how AI agents interact with business applications and data — language that points squarely at the SaaS-resident agent problem rather than the endpoint-resident one. Reco also disclosed100+ customers, with financial services accounting for roughly 40% of its installed base — the segment most exposed to model-risk, third-party-risk, and SEC/OCC governance overlays.

On top of the funding, Reco announced in late July that it had been named an OpenAI Select Partner for enterprise security, and in September shipped its first purpose-built AI Agent Security module for the OpenAI agent stack. The OpenAI partnership plus the AT&T round triangulates Reco's go-to-market: it is selling into the customer base that is actually deploying OpenAI agents, not the one thinking about doing so.

The bundled platforms are not standing still. Each has either shipped, acquired, or partnered an agent-runtime capability in 2026, and the M&A calendar is unusually dense for a category that, per Gartner, totals $2.84 billion in 2026 spending. The platforms' problem is that they started from endpoint, network, or identity — not SaaS-resident agent governance — and the gap between those starting points and where Reco sits is what the bundle math now has to clear.

The bundle thesis — one platform, one contract, one console — is strongest when the threat sits at a layer the platform already owns. AI agents threaten at three layers simultaneously: the endpoint (where the agent executes), the network (where it moves data), and the SaaS app (where it acts on records). The first two are where Palo Alto Networks, Zscaler, and CrowdStrike live. The third — the actual decision surface of an enterprise agent — is where the agent reads a Salesforce record, posts to a Workday workflow, or calls an OpenAI tool. That layer is invisible to endpoint and network controls.

Two pieces of evidence show the bundle math is straining. First, Gartner's August 2026 forecast separates the AI-security market into four subsegments — AI Application Security, AI Usage Control, AI Governance Platforms, and AI Gateway — and the fastest-growing segment is AI Usage Control (+73% YoY to $749M in 2027). That is the segment Reco's Browser Guard and the platform's runtime enforcement tools compete in, and it is growing faster than the AI Application Security layer where the platforms' main AI investments sit. Second, Okta — the platform closest to Reco's SaaS-resident thesis — felt compelled to give Agent SSO away free in core plans; the company is choosing adoption over ARR because letting Agent SSO sit behind a paid SKU would have ceded the entire agent-identity beachhead to Reco, SentinelOne, or Microsoft.

The numbers matter for stock-level theses on the platforms. Palo Alto Networks reported Q4 FY26 revenue of $3.41 billion (+34% YoY) and NGS ARR of $9.10 billion (+63% YoY), including Prisma AIRS at $100M+ ARR after only four quarters — the fastest product launch in company history, per its Q4 FY26 earnings presentation. That pace is the defense against Reco-style point tools, and it is real: a 280-customer Prisma AIRS installed base implies PANW has reached the same enterprise tier Reco is selling into, just via a different on-ramp. The risk is that the next 280 customers may already have Reco's Browser Guard installed at the browser layer, making PANW's runtime story additive rather than replacement. That is the bundle math that gets harder.

Zscaler took a different route. Rather than building an agent gateway from scratch, it paid for Symmetry Systems' access-graph technology in May 2026, and in Q4 FY26 its AI security bookings jumped more than 50% sequentially, with AI Protect crossing $100M in trailing-twelve-month bookings. The strategic bet is that AI agents should be governed at the network/data-access boundary — the same bet PANW's Console and Prisma AIRS 3.0 AI Agent Gateway make. If both bets are right, agents will be governed at the edge, not in the SaaS tenant — which would shrink Reco's beachhead. If the bet is wrong, Reco's 280-app discovery graph will look like the only map that matters.

CrowdStrike's position is the cleanest to read. Its Q2 FY27 print — revenue $1.47 billion (+26% YoY), ARR $5.84 billion (+25% YoY), net new ARR $332.8 million, and Falcon Flex module adoption at 51% (6+ modules) per the August 26, 2026 release — proves the bundle is being upsold aggressively. The agent-runtime answer is Falcon Guardian / Falcon AIDR, anchored on endpoint telemetry and extended in2026 to Copilot Studio and Claude Code. That works for agents running on the device. It does not work for agents running as a managed identity inside Salesforce or ServiceNow, where the agent has no endpoint presence at all.

Reco is the headline, but the bigger pattern is the category. Reco's $55M follows HiddenLayer's $100M Series B (Sep 2, 2026), Zenity's $125M Series C, Glow's $180M Series A at a $1.2B valuation (Jul 22, 2026), and Onyx's $113M round. Per Yahoo Finance's tally, venture investors deployed $435 million into 12 enterprise AI-agent security and governance financings between April and September 2026 alone. The deal sizes are now stretching into the territory where, instead of more venture rounds, an acquirer typically takes the next step — which is exactly the M&A calendar the bundled platforms are executing.

Reco's $55M is not a Series B — it is a referendum on whether AI-agent security is a feature or a category. The category side just won, by the only one that counts: $435 million of venture capital in five months, $1.45 billion of disclosed M&A in twelve, and a Gartner forecast that triples the spend by 2028. The platform side is not conceding — PANW's Prisma AIRS at $100M ARR in four quarters is the best counter-evidence — but it is now spending on the order of one acquired company per quarter to close the SaaS-resident-agent gap.

For investors, the practical question is whether the platforms' M&A strategy keeps them ahead of the discovery-graph moat that Reco, with 280 integrations, is building. The data point that will resolve it: how many Fortune 500 enterprises run Reco and a bundled platform's AI security product side-by-side through 2027, and which one survives the next round of consolidation. Right now, the check sizes — both venture and acquisition — suggest both sides think they will win.