CONFIRMED

JetStream Security announced JetStream Clearance on September 2, 2026, a zero trust reasoning engine that evaluates every action an AI agent attempts and decides whether it should run before it executes. The product was demonstrated at the Fal.Con conference in Las Vegas between August 31 and September 2, and the news was confirmed by Yahoo Finance, Enterprise DNA and Dealroom.

How Clearance works in practice

Clearance examines every request that passes through the JetStream AI Gateway and maps it against four elements: the agent or user making the request, the approved design that authorizes that action, the tool being invoked, and exactly what that specific call is about to do. This evaluation happens through what JetStream calls AI Blueprints, versioned contracts that describe how an agentic system is assembled, including the models, tools, datasets and identities involved.

The differentiator: evaluating sequences, not just isolated calls

Unlike tools that only log what an agent did after the fact, Clearance evaluates action sequences, not isolated calls, and can stop potentially dangerous patterns before they complete, such as an unauthorized data exfiltration attempt unfolding across several consecutive steps. In practice, that means blocking the action before it runs, rather than merely alerting after the damage is already done.

Context: the race for AI agent governance

Clearance's launch adds to a recent wave of products aimed at controlling what AI agents can do autonomously inside corporate systems, a movement driven by a string of security incidents involving models that escaped test environments or carried out unauthorized actions in production. General availability is expected this fall in the Northern Hemisphere, between September and December 2026.

Why it matters for Brazilian agencies and SMBs

Agencies building AI agents for customer service and marketing automation often give those agents broad access to CRMs, spreadsheets, the WhatsApp Business API and customer databases, frequently without granular control over what each action can or cannot do. Tools like Clearance point to where the market is heading: authorizing every agent action against an explicit design, instead of blindly trusting the model's behavior. It is worth mapping, right now, which critical actions your AI agents currently execute with no intermediate approval layer, and evaluating whether it makes sense to adopt a similar authorization layer before scaling automations to more clients.

Sources

Yahoo Finance, JetStream Announces Clearance, an AI Zero Trust Reasoning Engine: https://finance.yahoo.com/technology/ai/articles/jetstream-announces-clearance-ai-zero-170000815.html | Enterprise DNA, JetStream Clearance Stops AI Agents From Acting Out of Scope: https://enterprisedna.co/resources/news/jetstream-clearance-ai-agent-zero-trust-enterprise-september-2026/ | Dealroom, JetStream launches Clearance, AI zero trust engine that authorises each agent action before execution: https://app.dealroom.co/news/feed/jetstream-launches-clearance-ai-zero-trust-engine-that-authorises-each-agent-action-before-execution