Act Security Raises USD 60M to Put Boundaries Around AI Agents in the Cloud

Act Security emerged from stealth on July 28, 2026 with USD 60 million in total funding and a live platform that enforces deterministic access boundaries for humans, workloads and AI agents across cloud infrastructure. The capital arrives in two tranches: a USD 20 million seed led by Team8 and Bessemer Venture Partners, and a USD 40 million Series A led by Notable Capital. For enterprise security leaders, a new budget line just acquired a name: agentic access control.

Cloud access sprawl has been building for a decade. Enterprises layered cloud entitlement management, privileged access management and identity governance on top of each other, and the permission graph kept expanding anyway. According to the company, close to 97% of cloud access permissions sit dormant and unused, available for abuse the moment a credential leaks. AI agents turned that latent risk into an active one: they inherit standing human permissions and exercise them around the clock, at machine speed and machine scale. The founding team has already created one security category from an unmanaged asset class: CEO Jonathan Langer and CTO Itay Kirshenbaum built Medigate, the medical-device security specialist acquired by Claroty for USD 400 million. Bessemer backed that company too, which explains the fund's willingness to write a seed check while Act was a slide deck.

The timing tracks enterprise reality. Agent deployments moved from pilots to production across 2025 and 2026, and every deployed agent carries credentials: API keys, cloud roles, service-account tokens. Security teams that spent years shrinking human standing privilege now watch autonomous software accumulate it faster than any workforce ever did. Act's pitch lands on that exact gap, and it arrives from founders who have monetized an overlooked asset class before.

What changed: agentic access control gets institutional capital

The funding structure signals conviction on both sides of the table. Team8 and Bessemer led the USD 20 million seed, with Hetz Ventures and Claltech participating, at a stage when the product existed as an architecture on a whiteboard. Notable Capital then led the USD 40 million Series A, joined by Startpoint Capital and SVCI, and the platform now runs live in production environments. Four outlets covered the launch on the same day, a level of attention usually reserved for far larger rounds, which says as much about the category as about the company.

The product operates at the infrastructure layer. It evaluates identities, networks and AI access simultaneously, enforces least-privilege boundaries deterministically, validates those boundaries continuously, and blocks violations inside CI/CD pipelines before code reaches production. Compliance mapping covers NIST 800-53, PCI DSS and HIPAA, which shortens the audit conversation for regulated buyers. The company also positions the platform as an answer to the patch treadmill: AI tooling now finds fresh vulnerabilities faster than teams can remediate them, so Act removes the access paths that make those vulnerabilities exploitable in the first place. Langer frames the agent problem bluntly: "They're inheriting the same old human permissions, running around the clock at machine speed, with none of the judgment a person would apply." The leadership bench is complete from day one: Langer as CEO, Kirshenbaum as CTO, Stephan Goldberg as chief product officer and Ilai Fallach as VP of R&D.

What it means for the vendor map

Act lands in contested territory. Wiz and Orca own cloud posture, CyberArk owns privileged access, SailPoint and Okta own identity governance, and Palo Alto Networks sells across all three fronts. Every one of those vendors now faces a pointed question from customers: which product enforces what an AI agent can actually do at runtime, at the moment of action? Incumbents will answer with roadmap slides through the fall; Act answers today with a shipping platform and USD 60 million of runway. Expect two moves within twelve months: acquisition interest in agent-identity startups, and rapid feature releases that bolt agent scoping onto existing entitlement and PAM suites.

For buyers, the competitive pressure is welcome. A well-funded entrant pushes incumbents to price runtime agent enforcement as a feature rather than a new SKU, and it gives procurement a credible alternative to bring into renewal negotiations. The Medigate precedent matters here too: that team defined a category, proved it with enterprise deployments, and exited to a platform player at USD 400 million. Strategic acquirers, from Palo Alto Networks to CrowdStrike to Cisco, will read the same script, which means Act's independence has a clock on it. Buyers should reflect that in contract terms: portability clauses and data-export guarantees cost little today and protect you when consolidation arrives.

The 90-day decision

Run an agent-access audit before the end of October. Inventory every standing permission held by service accounts, workload identities and AI agents in your primary cloud estate, then measure the dormant share against Act's 97% benchmark; most teams that run this exercise land in the same range. Where agents inherit human roles, replace them with scoped, time-boxed credentials tied to named actions. Put one enforcement pilot in the CI/CD path so violations surface before deployment rather than after an incident. Then carry the results into Q4 renewals with your entitlement and PAM vendors: ask each to demonstrate runtime enforcement for agent actions on your own environment, and price Act against their answer. The enterprises that scope agent access this quarter set the terms; those that wait will negotiate under breach-response pressure.