Orca Security launches tools to find AI-built apps and agent blind spots in the cloud
Orca Security can now detect AI-built applications by their digital fingerprints and map autonomous agents' permissions, addressing software created outside formal security review.

Why it matters
- Orca Security announced tools on November 18, 2026 to detect AI-built applications and monitor autonomous AI agents in cloud environments.
- The system detects apps built with Lovable, v0, Replit, Base44 and Bolt, plus hosting services like Vercel, Netlify and Supabase.
- Agent monitoring covers Claude Code, Codex and Gemini CLI, with a new agents tab in Orca's AI Bill of Materials.
- Risk checks flag agents with excessive permissions, exposed MCP credentials, insecure HTTP connections and access to destructive tools.
- IDC Research Director Grace Trinidad said discovery must now start from what AI-generated apps leave in the environment, not builder declarations.
An employee can build and deploy an application over a weekend. Nobody in security reviews it. Nobody even knows it exists. That gap is exactly what Orca Security is now trying to close.
On November 18, 2026, Orca Security added tools to its platform that identify AI-built applications and monitor autonomous AI agents running in cloud environments. The update targets software created entirely outside standard security processes — apps built with consumer and developer AI tools that may already be running in production without formal review.
The stakes are straightforward. Business users now routinely build software with services like Lovable, Replit and Bolt, bypassing conventional engineering teams entirely. Developers, meanwhile, lean harder on autonomous coding agents that write code and carry out tasks on their own. Both trends expand what companies can ship. Both also expand what security teams can't see.
What can the new tools actually detect?
Orca's app discovery features hunt for digital traces that AI application builders leave behind. The system scans:
- DNS records linked to builder infrastructure
- Cloud workloads
- Code repositories
- Cloud identities and permissions
- Installed packages and file markers
- Commit identities
The detection list is specific. Orca says the platform can spot applications built with Lovable, v0, Replit, Base44 and Bolt, along with the hosting and data services those apps typically run on — Vercel, Netlify and Supabase. Findings surface through a service fingerprints view, asset-level details and a dedicated AI app generation dashboard.
The design philosophy matters here. Security teams usually know which platforms they've approved and monitor. What they don't know is which services employees used without registration or disclosure. Orca's approach flips the burden: instead of waiting for developers or business users to declare what they've built, the platform looks for evidence of activity already sitting in the environment.
"AI has turned everyone into a builder, but security teams are often the last to know what's been created," said Gil Geron, Chief Executive Officer and Co-Founder of Orca Security.
How does Orca monitor autonomous agents?
App discovery is only half the release. Orca has also expanded its monitoring of AI agents, covering both cloud-managed and self-hosted deployments. That includes coding agents such as Claude Code, Codex and Gemini CLI running on cloud workloads.
A new agents tab inside Orca's AI Bill of Materials gives security teams a central inventory of agents alongside the AI models, packages and MCP servers already tracked there. Users can filter that inventory by cloud, account and framework.
The platform then maps the relationships. It connects agents to the models they use, the tools and servers they touch, the cloud identities they operate under and the data stores they can reach. The goal is to let security teams judge how far an agent's reach actually extends — and where sensitive data might be exposed.
The risk checks are granular. Orca flags:
- Agents with excessive permissions
- Disabled approval controls
- Exposed credentials in MCP configurations
- Insecure HTTP connections
- Access to destructive tools
Those findings feed directly into Orca's existing risk scoring and attack path analysis, rather than living in a separate silo.
The release also covers AI-generated code itself. Organizations can attribute code to the specific coding agents and AI models that produced it. That gives security teams a view of how much AI-generated code exists across their repositories — and which tools contribute the most risk.
Why does this matter now?
The changes reflect a broader shift in how software gets made. AI tools have moved beyond engineering teams and into general business use, which means many more people can create and deploy applications quickly. Speed went up. Visibility went down.
For security teams, the problem isn't just that an unapproved application exists. It's whether that app can reach internal systems, cloud resources or sensitive corporate data. Autonomous agents compound the issue because they act using the permissions and credentials they inherit — an agent with broad access can touch sensitive data and take actions across cloud environments without any human in the loop.
Geron put the risk in concrete terms. "An employee can deploy an application over the weekend, while an autonomous agent can access sensitive data and take actions across cloud environments," he said. "Organisations need to understand not just what exists, but what it can access and where it creates risk. We're bringing that context to the entire AI development ecosystem so companies can keep innovating without losing control."
Industry analysts see the same problem. Grace Trinidad, Research Director at IDC, said discovery remains the central challenge — and that the old approach no longer works.
"Many organisations still struggle to discover what's been built and where it runs," Trinidad said. "Now that building is no longer confined to engineering, discovery has to start from what these AI-generated apps leave behind in the environment, not from the builder declaring anything."
What else changed in the platform?
Beyond the headline features, Orca is extending support in its AI application security tools to Base44, OpenAI and Vercel. It's also adding posture assessments for specific misconfigurations, including disabled row-level security in Supabase projects — a flaw that can expose database rows to users who should never see them.
The broader strategy is consolidation. Orca wants AI-built applications and autonomous agents pulled into the same cloud security workflows companies already use for vulnerabilities, misconfigurations, identities and sensitive data. Not a new dashboard bolted on the side. Same workflows, new objects.
That positioning matters for the market. Cloud security platforms compete on coverage, and AI-generated software is the fastest-growing category of untracked assets in most enterprises. The vendor that inventories shadow AI apps and agent permissions alongside traditional risks removes the excuse that AI development lives outside security's reach — and as coding agents proliferate, that inventory will only get longer.
Source: itbrief.com.au
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
708 articles
Related articles
- Trillium Labs Launches to Do High-Stakes AI Research in the Open
- Tanium launches AgentSight to track and govern AI tools on endpoints
- 87% of Enterprise Developers Use Low-Code. That Is the Agent Security Problem
- Islo Raises $15 Million for Controlled AI Software Factories
- 83% of Firms Run Unapproved AI Coding Agents, BlueFlag Research Finds