Nvidia launches security platform to restrain AI agents

By Splitfeed EditorialLast updated Oct 5, 2026, 16:30 UTC

In early October 2026, Nvidia unveiled a security platform built specifically to prevent AI agents from taking actions outside their defined parameters, a product framed around the growing concern that autonomous agents could cause harm if left unchecked. Within days of that announcement, the AI startup Manus moved in what appeared to be the opposite direction, equipping its agents with phone numbers and cryptocurrency wallets. Those capabilities give agents footholds in the real world, allowing them to make calls, receive verification codes, hold and transfer funds, and generally operate with less human intermediation than before. The two developments unfolded in the same week, with Manus's expansion reported on October 3 and a deeper analysis of Nvidia's platform published by SC World on October 5.

The juxtaposition matters because it illustrates a structural tension that goes beyond any single product launch. Nvidia is building guardrails; startups like Manus are building runways. The more consequential open question is a technical one: whether Nvidia's platform can revoke an agent's access after that agent has already been granted external capabilities such as a wallet or a verified phone number. If the answer is no, then containment tools may only be effective before deployment, not after, which would significantly limit their practical value in a landscape where capabilities are being extended rapidly and competitively.

No published reactions from the Pro-AI, Anti-AI, or Middle Ground camps were available at the time of writing, so what follows reflects what each camp would typically argue about a story like this. The Pro-AI camp would likely praise Nvidia's move as evidence that the industry is self-correcting, and view Manus's expansion as legitimate product development that increases agent usefulness. The Anti-AI camp would almost certainly point to Manus's timing as proof that voluntary or platform-level safety measures cannot keep pace with competitive pressure to expand agent autonomy, and would use the revocation question as evidence that containment is structurally inadequate. The Middle Ground camp would likely argue that both developments are necessary parts of the same ecosystem, that security infrastructure and expanded capabilities can and should evolve together, but that the revocation gap represents a real and unresolved engineering problem.

The most clarifying thing to watch is whether Nvidia or independent researchers publish a concrete answer to the revocation question, specifically whether the platform can claw back access rights for agents that have already established external accounts or credentials. That technical answer would either reinforce or substantially undercut the platform's core promise.