Enterprise AI shifting to managed deployments
Vendors and buyers are moving away from one‑off model access toward managed, workflow‑focused AI — OpenAI says enterprise now generates a large share of revenue and firms are adopting 'teams of agents' for repeatable processes. At the same time, model-makers are rolling out cybersecurity‑oriented tools cautiously and staging access to limit harm, which makes safety posture part of product‑market fit. (decrypt.co) (axios.com) (technobezz.com)
A year ago, many companies bought artificial intelligence the way they bought cloud storage: get access to a model, add a chatbot, hope employees find uses for it. In April 2026, OpenAI said enterprise customers already generate more than 40% of its revenue, and its pitch has shifted from single tools to company-wide systems of agents tied into real work. (openai.com) (decrypt.co) That changes what is being sold. Instead of paying for raw model access, buyers are paying for an intelligence layer that can sit inside claims processing, customer support, software development, and research workflows with permissions, internal data connections, and audit controls. (openai.com) OpenAI described the old problem in blunt terms on April 8: companies are tired of “point solutions” that do not talk to each other. Its answer is a unified operating layer, where multiple artificial intelligence coworkers share company context and can hand tasks off like a team on the same office floor. (openai.com) The revenue mix shows why vendors are chasing that model. Denise Dresser, OpenAI’s chief revenue officer, said enterprise is on track to reach parity with consumer revenue by the end of 2026, which means the biggest growth is no longer just millions of individual subscriptions but a smaller number of much larger deployments. (openai.com) (decrypt.co) OpenAI also gave a scale marker that sounds less like a chat app and more like back-end infrastructure. The company said its application programming interfaces now process more than 15 billion tokens per minute, and Codex has reached 3 million weekly active users. (openai.com) The cybersecurity push shows the next step in that enterprise shift. In February 2026, OpenAI launched Trusted Access for Cyber, an identity-based pilot that gives vetted defenders access to stronger cyber capabilities and set aside $10 million in application programming interface credits for the program. (openai.com) That pilot matters because the most valuable customers in security are not asking for a general chatbot. They want a model that can inspect code, trace attack paths, spot weak configurations, and plug into existing defense workflows without being opened to the whole internet. (openai.com) (tech.yahoo.com) Anthropic made the same bet this week, but even more dramatically. It launched Project Glasswing on April 7 and restricted Claude Mythos Preview to launch partners including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, Nvidia, and Palo Alto Networks, plus more than 40 other organizations tied to critical software infrastructure. (anthropic.com) (cnbc.com) Anthropic says Mythos can find and exploit software flaws at the level of top human security researchers, and reporting on the rollout says the model uncovered thousands of previously unknown vulnerabilities, including a 27-year-old OpenBSD bug and a flaw in FFmpeg that earlier automated scans had missed. (cnbc.com) (tech.yahoo.com) Once a model can both find a lock and explain how to pick it, safety stops being a policy document and becomes part of the product itself. Anthropic kept Mythos inside a locked-down consortium, and Axios reported that OpenAI is preparing its own restricted cybersecurity offering for a small group of partners rather than a broad public release. (axios.com) (tech.yahoo.com) That is the shape of the market now. The winning package is not just the smartest model, but the one that comes wrapped in identity checks, staged rollout rules, usage monitoring, and enough workflow plumbing that a bank, insurer, or software company can trust it with repeatable work every day. (openai.com 1) (openai.com 2)
