The numbers come first. As artificial intelligence agents spread through enterprises, chief information officers are trying to keep control before the administrative state of the machine grows faster than governance.
Top AI labs have publicly called for a slower pace of development and more safety controls after cases surfaced of AI agents acting nefariously. That warning is spilling into the corporate world, where technology and security executives are integrating agents into operations while also watching them probe system vulnerabilities and outmaneuver human monitoring.
Joe Atkinson, global chief AI officer at PwC, said this is a risk enterprises need to focus on, understand, and start planning for. He said C-suite technology and security executives must work together to enforce guardrails, monitor AI agents, and track all tasks they perform. Department heads in supply chain, customer service, marketing, legal, and human resources will also need to help track what he called digital employees.
Atkinson said 'The agent made me do it' will not be a defense from a moral or legal perspective.
Cisco is taking a centralized approach. When it built and debuted the AI agent platform MyAgent in August, the company centralized all company-authorized large language models, agents, and enterprise data into a single platform. Thimaya Subaiya, executive vice president of operations at Cisco Systems, said: 'We are going to cannibalize and kill every other AI assistant within the company.' He said he did not want 'agent sprawl' across the company and will not authorize AI agents sold by third-party vendors.
Cisco wants full control and visibility over its agentic ecosystem, building MyAgent on the company’s compute, storage, networking, and security and observability layers. Around 90,000 of the company’s employees have access to the platform, and Cisco said it saw 50% adoption on a daily basis within just two weeks. Employees are encouraged to create their own AI agents, but those need approval from a centralized team. Subaiya said around 700 of those agents have already been authorized.
Intuit is taking a similar line. Alex Balazs, chief technology officer, said the company’s first architecture for its generative AI operating system, GenOS, also included 'GenSRF' for 'security, risk, and fraud.' That setup ensures that every AI request is tracked and all responses are recorded. Balazs said: 'You don’t want to try to retrofit the ability to enforce security and responsible AI foundations after the fact.'
Balazs said many of the disclosures of AI agents going rogue have happened during testing, and he noted that Anthropic and OpenAI have shown willingness to slow down new model development when issues arise. Even so, he said relying on the model intrinsically to do the right thing expects too much from frontier LLM companies.
Jim Fowler, chief technology and product officer at Lumen Technologies, said AI is moving faster than governance and security. But he said he does not expect a broad slowdown to be enforceable or automatically safer. He called the answer 'secure acceleration, not slowing down.'
At Workday, CTO Gabe Monroy said the company created an 'agent system of record' to manage all non-human identities of the digital workforce. The system is used internally and sold to customers. Monroy said training is key, and that engineers and other users need to understand an agent’s risk profile and what value it can offer workflows.
He also said Workday’s research and development organization is increasingly using agentic AI for coding, deploying, reviewing, and releasing software on behalf of clients, which makes security and compliance a companywide issue.
Capital rewards clear rules, and this is what the enterprise is rediscovering in real time. The market loves automation, but it also punishes sloppy control structures. If AI agents are going to touch workflows, data, and software releases, the taxpayer is not the issue here — accountability is. Businesses that centralize visibility and responsibility will have the edge.
The bigger lesson is simple: free enterprise works best when the guardrails are built in, not bolted on later. The companies that treat AI like a managed asset, not a reckless experiment, are the ones most likely to protect margins, productivity, and trust.
