Obsidian Security

Somewhere in a Fortune 100 network right now, an OAuth token is being used by someone who shouldn't have it.

Nobody logged a failed password. No firewall blinked red. A legitimate credential, granted months ago to a marketing app nobody remembers approving, is quietly reaching into a system it was never meant to touch. This is the modern breach: polite, authenticated, and almost invisible. Obsidian Security exists to notice it anyway.

Obsidian sits where the enterprise actually lives in 2026 - not on laptops or in data centers, but inside the hundreds of SaaS applications, integrations, and now AI agents that run the business. Its platform watches roughly 49 million identities and 340,000 applications, and stops about 3,450 SaaS threats every month. It is, in the company's own framing, trying to become the confidence layer for an era when software increasingly acts on its own.

Pictured: a quiet network, moments before someone notices the thing that was always there.

The perimeter dissolved, and most security tools were still guarding the door.

For two decades, security meant defending endpoints and networks. Then the work moved. Email, code, customer data, finance, HR - all of it migrated into SaaS apps owned by other companies and accessed by anyone with a token. The attack surface stopped being a building you could lock. It became a sprawl of permissions, integrations, and forgotten admin grants spread across vendors nobody fully tracks.

That sprawl is the central tension of Obsidian's existence. Every convenience the SaaS era delivered - sign in with one click, connect this app to that one, let an assistant read your inbox - is also a door. Attackers stopped breaking in. They started logging in. And the tools built for laptops and firewalls simply could not see it.

Convenience and vulnerability, it turns out, are the same feature viewed from different sides.

Three people who had already built the last generation of security decided to build the next one.

In 2017, Glenn Chisholm, Ben Johnson, and Matt Wolff placed a wager that looked early at the time. Chisholm had been CTO of Cylance. Johnson had co-founded Carbon Black and, before that, worked as a practitioner inside the NSA and CIA. Wolff had been Cylance's chief data scientist, also with NSA roots. They had spent their careers defending endpoints. Their bet was that the endpoint no longer mattered as much as the identity using it.

Greylock led a $9.5 million Series A on the strength of resumes and a thesis. The thesis: SaaS would eat the enterprise, and someone would need to secure the apps from the inside rather than the network edge. It was the kind of bet that is either obvious or absurd, depending entirely on whether you are right. They were early. They were also right.

Three founders, two intelligence agencies, and a combined allergy to letting strangers log in unnoticed.

One platform that does the boring part and the scary part.

Obsidian's platform splits the SaaS problem into two halves. The boring half is posture: who has access to what, which settings drifted out of compliance, which app got connected to which other app while nobody was looking. That is SaaS Security Posture Management. The scary half is the live attack: a stolen token, a bypassed MFA prompt, a session quietly hijacked. That is Identity Threat Detection and Response, and it is where Obsidian made its name.

The trick is behavioral. Obsidian learns what normal looks like for each identity, then flags the moment a legitimate credential starts behaving like someone else is holding it. Lately the company has extended that same logic to a newer and stranger category - AI agents, which now reach into SaaS data with permissions of their own and no human watching the keyboard.

The product does two things at once: tells you the doors exist, then watches who walks through them.

The receipts: customers, numbers, and a growth curve that got attention.

A security thesis is only as good as the logos that bet their data on it. Obsidian's public references read like a directory of companies that cannot afford to be wrong about SaaS risk: Snowflake, T-Mobile, Databricks, Pure Storage, Seagate, Upwork, BigCommerce. The company says it serves 10 of the top 12 financial services firms, 3 of the top 5 healthcare firms, and 2 of the top 3 telecoms.

Numbers self-reported by the company. Big numbers, but the kind a CISO can actually check.

Forrester named the company a Strong Performer in its SSPM evaluation, with high marks for adoption. Partnerships extend the reach: the CrowdStrike Falcon integration pushes Obsidian's SaaS signals into the security operations center where analysts already live.

From "secure the apps" to "secure the things the apps are starting to do on their own."

The mission has quietly grown more ambitious. Obsidian started by securing SaaS applications. It now wants to be the trust layer for the entire AI era - the place where every application, agent, and integration is governed so that enterprises can, as the company puts it, move forward with AI fearlessly. Co-founder Matt Wolff's shift from chief scientist to Chief AI Officer is the tell. The frontier moved, and so did he.

There is a neat irony here. The same automation that makes AI agents useful - their ability to act without asking - is exactly what makes them dangerous in a SaaS environment full of standing permissions. Obsidian's pitch is that you do not have to choose between ambition and safety. You just have to watch carefully, which is the one thing the company has always done.

The mission grew, but the instinct stayed the same: assume the convenient thing is also the risky thing.

Back to that borrowed token.

Return to the opening scene. An OAuth token, granted long ago, reaching somewhere it shouldn't. In a network without Obsidian, that story ends weeks later with a forensic report and an apology. In a network with Obsidian, it ends in minutes - because the platform already knew what that identity normally did, noticed when it stopped behaving like itself, and pulled the thread before it became a breach.

That is the whole company in one sentence. Not a wall, not a lock, but a careful observer of the thousands of small permissions modern software hands out and forgets. As AI agents multiply those permissions faster than any human can track them, the job only gets bigger. Obsidian's bet from 2017 - that identity, not the endpoint, is the real frontier - looks less early every year. The token never announced itself. Someone still has to.