Airlock Digital Wants to Watch What Your AI Assistant Actually Does, Not Just Whether It's Allowed to Run
A new product layer from Airlock Digital tracks AI agents command by command, in real time, on the devices where they do their work.
Key points
- Airlock Digital announced Agentic AI Control & Governance at Black Hat USA 2026, targeting AI agents running on workplace computers.
- The Cloud Security Alliance reported in April 2026 that 82% of organisations had unknown AI agents running in their systems.
- 65% of those organisations experienced an AI-agent-related security incident in the 12 months before that April 2026 report.
- The product is expected to reach general availability in Q3 2026.
- Airlock Digital, founded in Australia in 2013, sells software that controls which programs are allowed to run on a company's computers.
Most workplace security software asks one question: is this program allowed to run? Airlock Digital says that's no longer enough.
At Black Hat USA 2026 in Las Vegas, the Australian firm announced Agentic AI Control & Governance. The product targets AI agents: software that acts on a user's behalf, making decisions autonomously rather than waiting for human instruction. A tool that can browse files or schedule meetings without being told to do each step individually.
Why does this matter to ordinary workers?
If your workplace uses AI tools, there is a real chance some are running without your IT department's full knowledge. The Cloud Security Alliance, an industry body focused on computing security, found in April 2026 that 82% of organisations had unknown AI agents operating inside their systems, and 65% had experienced an AI-agent-related security incident in the prior 12 months. We covered that data on 4 August in our report on AI-related incidents and API vulnerabilities.
Knowing a program is on your network is one thing. What it does once it starts is a different problem entirely.
How does the new product work?
Airlock Digital already sells application control software, a security tool that maintains an approved list of programs and blocks everything else. The new product extends that idea to cover what an approved AI agent does after it starts running.
Each command an AI agent attempts is logged, checked against policy in real time, and either permitted or blocked. When an action is blocked, the system communicates the boundary back to the agent so it can adapt rather than loop through the same rejected request repeatedly. That distinction matters: AI agents do not simply stop when refused.
"AI agents don't simply stop when an action is blocked; they evaluate alternatives and continue working toward their objective," said David Cottingham, Airlock Digital's co-founder and chief product officer.
A central dashboard records sessions, file interactions, token usage (the computing resource AI tools consume with each task), and cost.
| Feature area | What it does |
|---|---|
| Behaviour discovery | Logs AI agent commands and sessions automatically |
| Policy management | Central control with version history and admin permissions |
| Real-time evaluation | Checks each agent command against policy before it runs |
| Governance reporting | Searchable dashboard covering activity, risk, tokens, and cost |
| Availability | Customer general availability expected Q3 2026 |
This is the same gap that drew $113 million to Onyx Security in July, as we reported on 30 July. The pattern is consistent: vendors are converging on endpoint-level agent governance as the problem traditional security tools were never built to handle.
What should employees and managers do now?
Ask your IT team whether AI tools are inventoried and monitored. An unclear answer is itself useful information. You don't need to stop using the tools. You do need someone accountable for what they are doing on your systems. Policies governing which files, accounts, and external services an AI agent can touch are worth writing down before an incident forces the conversation.
