Altman warns firms to adopt AI now before rivals do at Cisco Summit

Sam Altman showed up at the Cisco AI Summit 2026 in San Francisco with a message that sounded less like a product pitch and more like an operating manual. Saying the next wave of AI will be decided by whether companies can run agents safely inside real workflows, not by whether the next model feels a little smarter.

Altman described AI moving past the chat box and into systems that can operate a computer the way a person does, hopping between tools, carrying context, and finishing multi-step work. He framed coding as the near-term wedge because it turns AI from advice into execution. That once an agent can write, run, and revise software in a loop, it starts changing how work gets done.

That forces a different set of problems as Altman kept returning to access and control because once AI is generating changes at machine speed, the failure mode is that "you can’t govern" what it produces properly.

Even a recent GitClear study states that codebases are showing more duplication and fewer refactoring signals as AI assistants spread, which is exactly the kind of drift that compounds when review and guardrails lag. 

And in that light, Altman points out that if companies cannot define what an agent is allowed to see, what it is allowed to change, what requires explicit approval, and how actions are reviewed, the rollout turns into a security debate that never ends.

He used the week’s viral agent chatter to underline the point. Altman said Moltbook, the viral distraction, may fade, but OpenClaw, the functional operator, is the real signal, and he flagged the tradeoff this creates. The moment an agent can act, it creates new dangers. When an agent has the keys to the cloud, a prompt injection is not a prank, it can become a data breach, a danger Wiz has warned about.

Altman’s bottom line was that AI is already crossing the threshold into doing work, but the world around it is behind. The advantage will go to organizations that rebuild the boring parts first, identity and permissions for agents, monitoring and audit trails, and the infrastructure to run these systems reliably at scale.