Every business deploying AI agents right now is quietly building an attack surface they haven’t fully counted yet. Obsidian Security just raised $85 million at a $1.1 billion valuation to make sure that doesn’t end badly for them.

The Palo Alto-based cybersecurity firm closed its Series D round led by Crescent Cove Advisors, with participation from existing investors Greylock Partners and Menlo Ventures. The raise reflects something the broader market is just starting to understand: AI agents aren’t just productivity tools. They’re identity holders.

The Ratio That Should Alarm Every Business Leader

Here’s the number that frames everything: inside the average enterprise, non-human identities outnumber human ones by 144 to 1.

Every API key, service account, and now every AI agent browsing Salesforce, querying Snowflake, or reading from GitHub on your behalf — each one is an identity with access rights, a potential entry point, and almost no oversight in traditional security tooling. That ratio existed before AI agents arrived. Agents have made it explode.

Obsidian’s platform tracks and governs these non-human identities across the third-party applications that run modern businesses: Google Drive, Notion, Slack, Salesforce, Snowflake, Databricks, GitHub, GitLab. When an AI agent connects to any of those tools, Obsidian sees what it accessed, what permissions it used, and whether any of that behaviour looks wrong.

Why This Funding Round Matters Now

This is not a theoretical risk. As enterprises have moved from AI pilots to production deployments in 2026, the attack surface created by agents operating inside SaaS applications has become very real. Obsidian’s customer growth tells that story: the company now counts more than 100 customers spending over $100,000 per year on its platform, with more than 14 spending above $1 million. Sixty Fortune 500 companies are in the base, spanning financial services, social media networks, and telecoms.

Those aren’t early-adopter vanity metrics. Those are the numbers of an enterprise security category that has found genuine product-market fit because the problem it solves is actively costing companies money and risk.

The timing also aligns with how the platforms themselves are changing. Anthropic’s Claude, OpenAI’s ChatGPT, and Microsoft Copilot Studio are all deepening their integrations with enterprise SaaS. Every new integration is another permission scope, another token, another session that security teams need to monitor. Obsidian is positioning itself as the governance layer that lets businesses accelerate agent adoption without losing visibility.

What This Means for Business

If you’re deploying AI agents — through a platform like Omni Ops, through Copilot Studio, through custom-built pipelines — there are three things worth addressing now rather than later:

Inventory your non-human identities. Most security teams can enumerate their human employees. Very few can produce a complete list of every API key, service account, and agent session active in their environment. That gap is what attackers exploit.

Treat agent permissions like employee permissions. The principle of least privilege applies to AI agents exactly as it does to people. An agent that only needs to read from one Salesforce object doesn’t need org-wide CRM access. Scoping matters.

Plan for agent-specific monitoring. Traditional endpoint security and identity tools were built for humans interacting with systems in human ways. AI agents operate at different speeds, in different patterns, and with different access sequences. Detection logic that catches suspicious human behaviour will miss a lot of what an agent does.

The market is signalling that this is a priority. Enterprise security buyers spent the first half of 2026 watching the agent ecosystem mature. The second half is where they start spending to secure what they built. Obsidian’s fundraise is one sign of where that money is going.

The Broader Picture

Obsidian’s unicorn round isn’t an isolated event. It sits alongside a wave of investment into AI agent infrastructure security: from MCP security tooling to runtime governance platforms to identity-first agent design. The common thread is that enterprises are learning, sometimes the hard way, that moving fast on AI deployment without equal investment in oversight creates compounding risk.

The businesses that will get the most from AI agents in the next 12 months aren’t the ones that deploy the most agents. They’re the ones that deploy agents they can actually see and control. Security is not a barrier to AI adoption — it’s the foundation that makes sustained AI adoption possible.

For organisations building an AI strategy, this is the moment to make security a first-class consideration alongside capability and cost. The investment is clearly being made at the infrastructure level. The question is whether your own deployment approach is keeping pace.