JetStream Security enables on-demand shutdown of compromised AI agents

JetStream Security has announced the release of an AI Kill Switch that allows organizations to shut down compromised AI agents on-demand without impacting other AI operations. This new control plane for AI agents solves the inability to stop a single agent that falters, begins overspending, or needs to be taken offline for compliance reasons without affecting other agents.

JetStream’s AI Kill Switch solves a problem many businesses have not faced up to this point: AI systems are becoming increasingly capable and as evidenced by the recent OpenAI/Hugging Face incident, can cause significant collateral damage as they seek to attain their goals in ways that we cannot predict.

This new reality means that every organization needs to have the ability to continuously monitor its AI systems and have the ability to precisely halt malicious actions using a kill switch.

“Not being able to stop a runaway AI system without impacting other production systems is one of the chief reasons companies hold back on AI,” said JetStream Security CEO, Raj Rajamani. “Only 17% of organizations are running AI at production scale, held back by a lack of trust and control rather than by the models. The AI Kill Switch™ failsafe frankly reduces our customers’ anxiety. If one system goes wrong, they stop that one and everything else keeps running.”

To provide this level of precision, JetStream factors together the individual who invoked the action, the agentic identity that carries it out, the intent of the system as inferred from its AI Blueprint and historic actions, and the individual product owner accountable for that agent. Access entitlements are mapped out in JetStream AI Blueprints so that every agentic action is traceable back to a responsible owner, enabling the Kill Switch to revoke exactly the right access and shut off the correct components in one pinpointed action.

Because each agent’s authority is carried by its own entitlements, the AI Kill Switch affects the single agent alone, and these actions are reversible. JetStream furthers the building of trust by recording every action made by every actor with a corresponding timestamp that gives organizations a defensible account for post-incident review.

That record matters beyond the incident itself. It gives companies something boards, auditors, and a growing body of AI regulation increasingly ask for: proof that a person can halt an AI system on demand, and clear evidence of who stopped it and why.

“Regulators and standards bodies have quietly converged on something most enterprises haven’t caught up to which is if you can’t stop an AI agent, you don’t control it. The EU AI Act now requires a way to bring a high-risk system to a safe halt, and both the NIST AI Risk Management Framework and ISO/IEC 42001 expect organizations to be able to disengage or shut down AI that operates outside its approved boundaries. A stop button is no longer a ‘nice-to-have’, it’s the line between AI you govern and AI that’s governing itself.”