OpenAI urges enterprise CISOs to adopt agentic AI tools for cybersecurity

The push for autonomous agents in cybersecurity forces enterprises to weigh immediate defensive benefits against significant liability and safety risks. While Brockman frames this as a necessary response to evolving threats, the lack of explicit 'undo buttons' or audit trails for rogue agents remains a critical technical gap. For the broader ecosystem, this shift suggests that AI leaders are prioritizing revenue-generating security features over the rigorous safety guardrails established in earlier development phases. As OpenAI moves toward a potential public offering, its focus on defensive utility serves to reassure investors of its enterprise maturity. Watch for whether CISOs demand formal liability frameworks before granting agents deep access to high-priority infrastructure.

OpenAI's push for agentic AI in enterprise cybersecurity arrives as multiple sectors race to deploy autonomous agents in production environments. In the telecom vertical, Ericsson published a detailed framework in July 2025 describing agentic AI as the pathway to autonomous network level 5, claiming an 80 percent reduction in time spent on analysis and decision-making processes when agents handle routine network management and troubleshooting tasks. That same momentum is visible in enterprise networking, where Blue Planet and Telefónica Deutschland completed a joint proof of concept using agentic AI to power 5G network slicing, reducing slice design tasks from weeks to minutes by offloading complex standards compliance to AI agents operating within a multi-domain service orchestration environment. Recent data suggests that enterprise agentic AI adoption hits 59% as production deployments scale across industries. The business case for agentic AI in security and operations is being shaped by competitive pressure and market forecasts. ABI Research, cited in the Blue Planet coverage, forecast network slicing to become a $19.5 billion market by 2028, though earlier projections of $66 billion by 2026 have been revised downward as technical and use-case challenges slowed adoption. Ericsson's networks chief Per Narvinger framed AI-driven RAN optimization in similar economic terms at MWC 2026, noting that AI models can extract 10 percent more value from spectrum that operators have already optimized deterministically for 30 years, translating directly into billions of dollars given recent spectrum acquisitions. These economics mirror the argument Brockman makes for cybersecurity: that the cost of not deploying autonomous agents exceeds the risk of doing so. On the technical front, Ericsson demonstrated at MWC 2026 that its AI-native link adaptation technology, first field-tested by Bell Canada in April 2025, integrates AI into sub-components of the radio algorithm rather than as an add-on, executing spectrum optimization on the baseband unit in real time. The company also unveiled new radios featuring custom Ericsson Silicon chips with neural accelerators designed for on-site AI inference in massive MIMO systems. In the enterprise 5G router space, Ericsson claimed it achieved the first enterprise 5G agentic AI agent deployment, simplifying private network deployment and lowering barriers to adoption. These deployments illustrate the broader pattern OpenAI is betting on: that enterprises across sectors will accept autonomous agent access to critical systems once measurable efficiency gains are demonstrated, even as AI agent population safety risks remain undefined.