Act Security: who they are, what they sell and what CISOs should ask

What Act Security sells

Act Security sells a cloud security platform that removes excess access from cloud infrastructure. It is aimed at security teams and, through a free toolkit, at the engineers who write cloud access policies. The company is based in Tel Aviv, per its stealth-exit announcement, and was founded in 2025, according to a startup directory profile. It is privately held and venture-funded. A company release states that it "has raised $60 million from investors, including Team8, Bessemer Venture Partners, and Notable Capital". The company says it came out of stealth on July 28, 2026, with that sum split across a $20M seed and a $40M Series A.

Its platform page says Act "eliminates unmanaged access at the infrastructure layer". The page states that identity, network and AI access are analysed and enforced as a single model, and that Act places six boundaries across a customer's cloud. The work is organised into hardening campaigns of two kinds:

• Guardrails, which the page describes as "Broad explicit-deny policies that block entire attack categories".
• Ring-fencing, described as least-privilege allow policies "applied selectively to crown jewel applications and sensitive data."

The company says each recommended change is simulated against historical access data before enforcement. It also says humans make every decision, with agentic workflows assisting.

On delivery, the platform page says enforcement "ships as pull requests to your existing Infrastructure-as-Code (Terraform, CloudFormation)" and promises "No agents and no new infrastructure to maintain." A HUB International case study published by the company described Act as connected read-only to the customer account. The case study said it proposes narrowed policies that the customer reviews and applies, and that "Act has no write path into the environment, by design." The trust center lists AWS as host for the platform and production environment. It also lists Anthropic and OpenAI language models as integrated into the platform.

The second line is Amphi, a free, open-source toolkit for cloud engineers. Act launched it one month after leaving stealth, the company says, and built it on Cloud Copilot, an AWS IAM tool it acquired. Its launch post says every tool runs in the browser, as an offline CLI or from source on GitHub, with no login or email required.

The company draws one line around what it is not: a findings tool. The platform page contrasts Act with tools that "stop at surfacing findings". The release says that "Rather than surfacing findings, Act eliminates the conditions that make them exploitable". Our record files the company under cloud identity entitlements, non-human identity security and zero trust network access. Its neighbours are listed on /vendors.

The founders

Act Security has four co-founders: Jonathan Langer (CEO), Stephan Goldberg (CPO), Itay Kirshenbaum (CTO) and Ilai Fallach (VP R&D). Langer is the one who speaks for the company in public. Its award announcement quoted him as CEO and co-founder.

The four share a history at Medigate. Langer and Kirshenbaum co-founded Medigate, a medical device security and management startup, in 2017, and Goldberg joined as one of its first employees, CTech reported. The same report said Medigate built a security platform for medical devices used in hospitals and healthcare institutions, and that Claroty acquired it in 2021 for approximately $400 million.

Act's own account of the team is broader. Its about page says the company "was founded by a team of builders and operators with decades of shared experience across security and infrastructure, including healthcare and critical infrastructure." The page names Medigate as part of that experience and states that the team helped "hundreds of organizations eliminate risk at its foundation" there.

The same page gives the company's stated reason for starting Act. It says the Medigate years produced a belief that "lasting security addresses risk at its root", and that "We built Act to bring that same root-cause approach to the cloud."

Langer has framed the problem in similar terms. In the October 2026 award announcement he said: "For a decade, the industry obsessed over chasing vulnerabilities and findings, while cloud access quietly sprawled past anyone's ability to track it." He added that "AI took away the luxury of time." On AI agents in particular, he was quoted at the July launch saying "They're inheriting the same old human permissions, running around the clock at machine speed, with none of the judgment a person would apply."

He has also said who he thinks does the work. When Act announced Amphi in September 2026, Langer said "Every CISO we talk to already wants to fix access sprawl, but it's their infrastructure and platform teams who actually have to manage the access behind that decision." He gave that as the reason for buying Cloud Copilot, saying the company did it "to support the community that actually operates the cloud."

The founding team also works directly with customers. The company's write-up of a two-day workshop with HUB International in September 2026 lists Act's founding engineers among the people in the room, alongside HUB's teams and AWS staff. More on people who run security is on our people pages.

What the product does

Act describes its main product as a cloud access security platform that works in four stages, each set out on its platform page. The first stage maps the environment. The company says Act "systematically maps everything running in your cloud, whether you know about it or not," and organises the map around real applications, environments and data.

The second stage models reachability. Act's page states that it brings "identity, network, and AI access controls together into one unified access map," then compares three views of that map: intended access, observed access and configured access. The company's argument is that "the gap between what is permitted and what is actually required is where risk accumulates."

That model is organised around six boundaries the company names: human identity, environment, application, AI, data and external. Each describes a target state. For AI, the page says Act will "ring-fence cloud AI agents so they can't reach beyond their intended scope". For data, it cites geo-fencing for GDPR and limits on regulated data to meet NIST 800-53, PCI-DSS and HIPAA.

The third stage, the hardening campaign, is where a security team does most of its work. Act curates the policy changes and groups them into two campaign types. Guardrails are broad explicit-deny policies aimed at whole attack categories, such as cross-environment movement, data exfiltration and external overexposure. Ring-fencing applies explicit-allow, least-privilege policies selectively to crown-jewel applications and sensitive data. The company says a single policy update can eliminate thousands of attack paths. At the HUB International workshop mentioned earlier, Act presented an "Adopt AI safely" campaign. It was a set of controls running from execution-role scoping to account guardrails, and each was an item HUB could review and apply through its own infrastructure-as-code.

The fourth stage is enforcement. Before a change ships, Act simulates it against historical access data to show what it would have blocked. The page says "humans lead every decision, assisted by agentic workflows". It also says enforcement continues as the environment drifts, from CI/CD pipelines to runtime. Changes reach the cloud through the pull-request route described above, so the team keeps the approval step.

Alongside the paid platform sits Amphi, the free toolkit, built from seven single-purpose tools:

• IAM Expand takes a policy, a code snippet or a single action and expands each wildcard into the actions behind it.
• IAM Truth renders an SCP or RCP as a truth table of conditions and outcomes.
• IAM Simulate evaluates a request locally and explains the decision.
• IAM Shrink compresses an action list to fit AWS character limits without changing what it grants.
• IAM Convert generates Terraform, CDK or CloudFormation from a JSON policy.
• IAM Collect and IAM Lens work as a pair: the first pulls live policies into a local dataset and the second queries that dataset for reachability.

On our reading, the free tools cover analysis of policies and accounts, while campaigns and the enforcement loop stay in the paid platform.

Figure: Act Security's product page, captured 2026-10-29. Source: Platform | Act Security.

How it works and what it connects to

Act's platform page says the product works through controls the customer already has. Its page states that Act enforces boundaries "through your cloud's own native controls, using Infrastructure-as-Code", with "No agents and no new infrastructure to maintain." Recommended changes are delivered as pull requests to existing Terraform or CloudFormation. The company's award announcement describes the boundaries as drawn across multi-cloud environments and enforced at the infrastructure layer. A company blog post lists the native controls it means: "Security Groups, NSGs, SCPs, RCPs, guardrails, private endpoints etc."

The most detailed account of a live connection comes from the HUB International case study the company published. In that deployment, HUB's own team set up cross-account read-only access for Act, inside HUB's change-control rules. The study says that once connected read-only, Act compares configured access against observed access for every runtime and execution role, and proposes the narrowed policy. It also states a limit plainly: "HUB reviews, HUB applies. Act has no write path into the environment, by design." Our reading is that enforcement in that setup depends on the customer merging the changes. The case study does not say whether every customer is connected this way.

On where customer data lives, the company's trust center lists Amazon Web Services as the host for the Act Security platform and production environment, in the United States (region per customer preference) and EU / UK. Customer login to the platform runs through auth0 by Okta. The same page lists ISO/IEC 27001:2022 and SOC 2 Type 2 under compliance.

The trust center also shows language models inside the product. It lists Anthropic for "AI language model services (Claude) integrated into the Act Security platform" and OpenAI for the same role with GPT. It lists Nebius for "AI/ML model inference and GPU compute infrastructure". The sources do not describe what these models do in the analysis, or whether customer configuration data is sent to them.

The free Amphi toolkit, mentioned earlier, runs on the user's machine. The launch release says each tool ships as a CLI that runs fully offline and a browser version that runs entirely on client-side code. Its action data comes from iam-data, which refreshes daily.

The sources do not publish a list of supported cloud platforms for the commercial product or name integrations with identity providers, ticketing or SIEM tools.

Where it sits in the market

CISO Tribune's record files Act Security under three categories. Cloud identity entitlements (CIEM) covers the analysis and reduction of the permissions granted to human and machine identities in cloud platforms. Non-human identity security covers the discovery, governance and protection of service accounts, API keys, tokens and workload or AI agent identities. Zero trust network access (ZTNA) covers standalone services that grant users access to specific private applications after checking identity and device, in place of a network-level VPN.

The chart shows how many vendors CISO Tribune's record files under each of Act Security's three categories.

Other vendors on the record in those categories, in alphabetical order, are 1Password, Aembit, Akeyless, Apono, Appgate, Aryaka, Astrix Security, Auth0, BeyondTrust, Britive, Cato Networks, ConductorOne, CyberArk, Cyera, Cygrid, Cyscale, Delinea, Descope, Dispel, Dispersive Stealth Networking, GitGuardian, Grip Security, HashiCorp, Infisical, Ivanti, Juniper Networks, Keeper Security, Netskope, Oasis Security, Onyx Security, P0 Security, Permiso, Ping Identity, Reco, SailPoint, Saviynt, Silverfort, Sonrai, SSH Communications Security, Symmetry Systems, Token Security, Truffle Security, Twingate, Valence, Veza, Xage Security, Xcitium, Zero Networks and Zscaler. Their profiles sit in the vendor directory. A shared category does not mean two products do the same job.

Act's own pages draw the line against adjacent tools mainly by what happens after a finding. The platform page states that "Most cloud security tools stop at surfacing findings, leaving teams to chase endless issues." Its stated alternative is that "Act removes excessive cloud access, neutralizing vulnerabilities at the source, before they can be exploited." The same page argues against splitting the problem across products: identity, network and AI access, it says, "are analyzed and enforced as a single model, closing gaps that siloed tools cannot structurally reach." A company blog post made the same case, saying identity and network "just landed in separate tools, and neither one tells the full story of access in the cloud."

On zero trust, Act announced that it had been selected as "Zero Trust Network Access Solution of the Year" in the CyberSecurity Breakthrough Awards. That announcement, beyond the infrastructure-layer boundaries already described, placed Act's enforcement "past the point of access where most zero-trust tools stop." In our reading, that puts Act at a different point from the user-to-application brokering in the ZTNA definition above: its pages describe restricting what identities can reach inside the cloud, not replacing a VPN.

The identity categories map onto two of the six boundaries on the platform page. For people, the stated aim is to "Rightsize human access and remove standing privileged access, so nobody carries more rights than their role actually needs." For machines, the AI boundary aims to "Contain endpoint AI agents, govern the infrastructure they run on, and ring-fence cloud AI agents so they can't reach beyond their intended scope." Terms used here are defined in the glossary.

Funding, investors and growth

Act Security raised all of its disclosed money before it showed itself publicly. Its own news page states that the company "emerged from stealth on July 28, 2026 with $60M across a $20M seed and $40M Series A" (Act Security). The same page describes the company as a Tel Aviv startup. Both rounds had closed months before the stealth exit.

Seed. The $20 million seed round was first reported in August 2025. CTech later wrote that "The company’s Seed round was first revealed by Calcalist in August" (CTech). According to the same report, the investors that led the seed were "Bessemer Venture Partners, Team8, Hetz Ventures, and Brightmind" (CTech).

That brought its total to $60 million. The report named the participants:

• U.S.-based Notable Capital led the round.
• Startpoint Capital, a new fund led by Sharin Fisher Dibrov, participated.
• Claltech participated.
• The seed investors returned.

In its own announcements the company names a shorter list. A September 2026 press release states that "Act has raised $60 million from investors, including Team8, Bessemer Venture Partners, and Notable Capital" (PR Newswire).

Valuation and headcount. Neither the company nor the funding report states a valuation. The only headcount figure comes from the Series A report. CTech wrote that the company "currently employs 30 people" and described its work as a platform "for securing cloud infrastructure and data centers" (CTech). That count is from December 2025 and is likely out of date ten months later, but that is our reading, not a reported figure.

What the money was for. None of these sources sets out how the funds would be spent. The nearest statement is the company's framing at launch: it said its platform "targets AI agents inheriting bloated cloud permissions" (Act Security).

First acquisition. The first visible use of the company's resources after launch was a purchase. It used that AWS access-control technology as the base for its free Amphi toolkit. The sources give no price or terms for the deal.

The acquisition also brought in Cloud Copilot's creator. The launch release quotes David Kerber as "Cloud Software Architect at Act and the creator of Cloud Copilot" (PR Newswire).

Other security vendors' funding and people changes are tracked under /vendors and /moves.

Customers, partners and public evidence

Act's public customer evidence so far comes almost entirely from its own site. Its platform page carries a block of endorsements headed "Hear from those who Acted" from security heads at named companies. They include Daniel Chen, CISO of Papaya; Chad Kalmes, CISO and CIO of Benchling; Bill Dougherty, CISO of Omada Health; Daniel Bren, global CISO of Armis; and Brian Kerr, CISO of Klaviyo. Only some of the quotes describe use directly. The page does not say which of the companies are paying customers.

The one detailed case study is HUB International, an insurance broker headquartered in Chicago, which Act wrote up on its blog. The post says HUB's developers already used Claude Code and that the company had picked Amazon Bedrock AgentCore for its AI build. According to the post, in September 2026 HUB's TechOps organization, AWS and Act spent two days in one conference room building a sandboxed environment for coding agents. The environment was deployed as modular Terraform into HUB's own AWS account. Monsef Elalamy, senior director of cloud and infrastructure at HUB, is quoted: "Two days, hands on keyboard, and we got it done." The post describes Act's continuing role as read-only.

It also includes entries for application penetration testing and responsible disclosure.

The company's community footprint is the Amphi toolkit. The Amphi page states that every tool is open source on GitHub and invites issues and pull requests. The launch post lists packages published under the @actsecurity name on npm. In our reading, that gives engineers a way to inspect Act's IAM tooling for themselves, which is the closest thing to checkable public evidence in these sources.

The sources show no partner program, cloud marketplace listing or third-party benchmark. Beyond the award covered earlier, the main figure Act puts forward is its own. At launch it argued that 97% of cloud access sits unused. That is the company's claim, and none of the sources given here test it.

Research, open source and public work

Act's public work so far falls into two groups: the Amphi toolkit and a run of blog posts written by its staff. None of the sources show a benchmark, a conference talk or a vulnerability that Act itself found.

Amphi. The toolkit launched on September 2, 2026. The release said it "covers the full policy lifecycle, from reading, testing, simulating, shrinking, converting policies, and querying reachability across the cloud." Act's launch post says it was "Built on the Unix philosophy of doing one thing well". Each tool covers one step:

• IAM Expand takes its action data from iam-data, which the post says refreshes daily.
• IAM Truth renders an SCP or RCP as a truth table, showing each combination of conditions and the allow or deny it produces.
• IAM Simulate runs locally against a specific request context and explains how the decision was reached.
• IAM Shrink works against AWS limits the post gives as 6,144 characters for a customer managed policy and 10,240 for a role's inline policies.
• IAM Convert generates Terraform, CDK or CloudFormation, according to the Amphi page.
• IAM Collect pulls live policies from AWS accounts into a local dataset, and IAM Lens queries that dataset to model reachability.

The release said every tool ships three ways: an offline CLI, a browser version running on client-side code, and full source on GitHub, alongside the repositories and packages noted earlier. The Amphi page also offers 30-minute sessions with Act engineers on a specific policy or access design.

Writing. On June 29, 2026, Offir Levy, VP Sales Engineering, published a post arguing that a cloud identity has to pass two tests at once, network reachability and IAM permission, and that enterprises check them separately. The post leans on outside figures rather than Act data. It cites Verizon's 2025 DBIR on stolen credentials as the most common way in, and a CrowdStrike measurement of a 29-minute average breakout time in 2025, with the fastest at 27 seconds.

On October 22, 2026, Kobi Rubin, VP Research, wrote about PHP2Shell, which the post describes as a pair of critical CVEs in core PHP that let an unauthenticated attacker run code or read credentials from memory. Its argument was that teams can remove the access an attacker would use even where they cannot patch every runtime at once. On our reading of the summary, it is commentary on the bugs rather than a disclosure by Act.

On October 26, 2026, Hadar Landau followed with a post arguing that AI is finding flaws faster than teams can patch them.

What a CISO should ask before buying

Our reading: these are the questions we would put to Act Security in a proof of concept. Each is tied to something the company says about its own product.

1. Run the simulation on your own history. The platform page says each recommended change is "simulated against historical access data, to show what it would have blocked". How far back does that history go? What happens to access used less often than that window, such as quarterly jobs or break-glass roles?

2. Trace every exception. The company describes guardrails as explicit-deny policies that keep legitimate access "through built-in exceptions". Ask who writes each exception, who reviews it, and how it expires.

3. Define intent. The page says decisions are weighed against "intended usage patterns". Ask what inputs define intent and whether your team can correct them.

4. Recount the paths. The company says one campaign can "eliminate thousands of attack paths with a single policy update". Ask how a path is counted. Then have it recounted on your estate before and after one change.

5. Read the pull requests. Enforcement, the page states, "ships as pull requests" to existing Terraform or CloudFormation. Ask how it handles resources not under code, and drift between the repository and the live account.

6. Confirm the access model. In the HUB International case study, the company says it connected read-only and "has no write path into the environment, by design." Ask whether that is the default for every customer, and what permissions the connector role holds.

7. Map the footprint. The page says "No agents and no new infrastructure to maintain." Ask what the deployment needs instead: roles, log sources, network reach.

8. Ask what reaches the models. The trust center lists Anthropic and OpenAI as subprocessors whose models are "integrated into the Act Security platform." Ask which policies, logs or metadata are sent, to which region, and whether this can be switched off.

9. Test the multi-cloud claim. A company release said Act "draws adaptive boundaries across multi-cloud environments." The open-source suite builds on an AWS IAM tool, and our reading is that its public work so far centres on AWS. If you run Azure or Google Cloud, ask to see enforcement there.

10. Scope a real agent. The page says Act will "ring-fence cloud AI agents so they can't reach beyond their intended scope". Pick one agent role in your environment and ask to see the narrowed policy and what it would have blocked.

11. Read the audit reports. The trust center lists ISO/IEC 27001:2022 and SOC 2 Type 2. Ask for both reports and check their scope and period.

Timeline

Act Security's dated record is short. The company was founded in 2025 by the Medigate team, and it did not present itself publicly until mid-2026, so most entries below fall between July and October 2026. The list runs oldest first. Each line comes from the company's own releases and blog or from press coverage, and where a source gave only a month or a year, the line gives only that.

The chart counts the company milestones cited in this profile by the quarter in which they fell.

Chart data: Milestones cited in this profile; most from TechLeap Tech map | Netherlands Startup map: TechLeap Tech map | Netherlands Startup map | ACT company information, .

• 2017: Jonathan Langer and Itay Kirshenbaum co-founded Medigate, a medical device security startup, CTech reported. Stephan Goldberg joined as one of its first employees.
• 2021: Claroty acquired Medigate for about $400 million, according to CTech.
• 2025: Act Security was founded in Tel Aviv by the former Medigate founding team, per its TechLeap listing.
• August 2025: The company closed a $20 million seed round, which Calcalist first reported that month.
• December 2025: CTech reported a $40 million Series A led by Notable Capital, four months after the seed. At the time the company had 30 staff.
• July 28, 2026: Act emerged from stealth with $60 million across the two rounds, the company's announcement states.
• September 2, 2026: The company announced the Amphi toolkit described earlier. Its own post called the Cloud Copilot purchase its first deal, made one month after leaving stealth.
• September 2026: HUB International's TechOps team, AWS and Act spent two days in one room building the AgentCore environment covered in the customer section.
• October 8, 2026: Act announced that the CyberSecurity Breakthrough Awards program had named it "Zero Trust Network Access Solution of the Year".
• Late October 2026: Act published the two research posts noted earlier, on PHP2Shell and on patching.

In our reading, the pace stands out: the sources place a stealth exit, an acquisition, a public toolkit and a joint customer build within roughly ten weeks. A buyer tracking the company should treat this as the record of a first public year, not a long operating history.