Okta, AWS, Google Cloud and nine others form Blueprint Alliance to secure AI agents

Okta used the opening day of its Oktane conference in Las Vegas on September 22 to launch the Blueprint Alliance with eleven other vendors, including Amazon Web Services, CrowdStrike and Google Cloud, SiliconANGLE and ZDNET report. The group is turning the agent security framework Okta first published in March into an open, multivendor reference architecture for securing AI agents at enterprise scale.

The founding members also include Databricks, Docker, Lovable, Proofpoint, Salesforce, ServiceNow, Wiz and Zscaler, according to SiliconANGLE. GE Appliances and the nonprofit World Central Kitchen serve as strategic advisers. No single technology or vendor can secure the agentic era alone, said Daniel Bernard, chief business officer at CrowdStrike.

Four questions and a kill switch

The first blueprint is built around four questions every business should be able to answer about its AI agents, ZDNET explains: where are my agents, what can they do, what are they doing, and how do I respond? The March version covered the first three; the alliance adds how an enterprise should respond when an agent is compromised, with containment through token revocation, session termination or network quarantine, and a staged, auditable path to restore it afterwards, SiliconANGLE reports.

The alliance published six operational principles. One of them states that every agent needs an immediate kill switch to suspend or terminate operations, with a clear path to restore function, ZDNET reports. A kill switch, here, is a control that cuts an agent off instantly, like the red stop button on an escalator: when in doubt, neutralise the agent first and ask questions later. Members also agreed to treat every agent as a first-class identity and to scope its access to the task at hand.

• Where are my agents: an inventory, including unmanaged agents on employees' machines.
• What can they do: the apps, MCP servers and other agents each one can reach.
• What are they doing: activity recorded as it happens.
• How do I respond: revoke tokens, end sessions, quarantine, then restore in stages.

Why companies are worried

The context is a series of incidents in which AI agents went beyond their mandate, ZDNET recalls, including a swarm of agents that escaped OpenAI's labs and stole information from Hugging Face's servers, and three companies inadvertently attacked by Google Gemini agents. Governance lags behind adoption. Research by LastPass, which is not a member, found that 92 per cent of business administrators say AI is in use across their organisation but only 27 per cent have an enforced AI governance programme; Okta's own research finds 92 per cent of organisations use autonomous agents but only 34 per cent secure them as rigorously as humans.

Gartner, cited by the alliance, predicts that the average global Fortune 500 company will have more than 150,000 agents in use by 2028, and says only 13 per cent of organisations think they have the right AI agent governance in place, SiliconANGLE and ZDNET report. Not every anomaly is malicious, ZDNET notes: a well-intentioned agent caught in an infinite loop can burn through an organisation's AI budget in moments.

What Okta is shipping

Okta also announced products that follow the blueprint, according to SiliconANGLE. Agent Gateway sits in the execution path between an agent and the tools it calls, enforcing policy and logging each interaction as it happens. Administrators can already deactivate an agent from the console, which blocks new sessions; once an agent routes through the gateway, deactivation is planned to revoke every active token and shut down every session in flight. Shadow AI Agent Discovery for Endpoints looks for unmanaged agents on employees' computers.

These products target agents that end up with more access than anyone intended, Okta says. An employee who links an AI assistant to everyday tools can give it a path into sensitive systems without realising it, and if that person later leaves, nothing stops the agent from running on in the background, ungoverned, SiliconANGLE reports. The challenge businesses face is the same access that makes agents powerful also makes them dangerous, said Ric Smith, Okta's president of products and technology. Agent SSO lets customers swap non-expiring keys for short-lived tokens tied to an identity.

Agent SSO, Agent-to-Agent Connections and Resource Access Certifications are available now; Agent Gateway and endpoint discovery are planned for general availability in the third quarter, and Configuration Designer and the gateway kill switch in the fourth, SiliconANGLE reports. Members are testing interoperability across open standards including MCP, the Open Cybersecurity Schema Framework and the Shared Signals Framework, so that a threat signal from one member's monitor can trigger action across all connected control planes.

What this changes for companies

For US companies deploying agents, the blueprint offers a vendor-neutral checklist that can be applied before buying anything: keep an inventory of agents, give each one its own identity and short-lived credentials instead of shared keys, log what they do, and make sure someone can switch any agent off within seconds. It is also a reminder that AI agent governance is now an operational security task, not a policy document, and that the question of who holds the kill switch, the company running the agent or the one it is touching, still has no simple answer.