Новини

OpenAI Acquires Ona to Enable Secure Enterprise Deployment of Codex Agents

· Forbes

OpenAI has announced its acquisition of Ona, a German startup formerly known as Gitpod, in a strategic move to address enterprise security concerns around autonomous AI agents. The deal integrates Ona's secure cloud execution platform with OpenAI Codex, enabling coding agents to operate within enterprise-controlled environments rather than on OpenAI's servers. The acquisition addresses a critical enterprise adoption barrier: governance and security. While Codex has grown to serve over 5 million weekly users with 400% growth since early 2026, enterprise customers remain hesitant to allow autonomous agents access to sensitive systems and data. Ona's platform provides the solution through what it calls "customer-controlled execution," where OpenAI supplies the AI model while the agent runs within the customer's own cloud infrastructure. Ona's technology stack consists of three core components: sandboxed cloud workspaces that deploy consistently, background agents that handle extended tasks, and comprehensive security guardrails including audit trails, role-based access controls, and deployment within customer virtual private clouds. This architecture allows enterprises to maintain full control over their data and security policies while leveraging AI capabilities. The timing reflects intensifying competition in the enterprise AI market. OpenAI faces direct competition from Anthropic's Claude Code, which has gained significant traction among engineering teams. Both companies have filed confidential SEC prospectuses, with OpenAI valued at approximately $852 billion, creating pressure to convert AI enthusiasm into production revenue. This acquisition follows OpenAI's pattern of strategic enterprise-focused purchases, including security testing startup Promptfoo and healthcare technology firm Torch. These deals prioritize enterprise infrastructure capabilities over raw AI model improvements, signaling OpenAI's commitment to enterprise market penetration. However, the acquisition raises important questions about model neutrality. As an independent vendor, Ona previously supported multiple AI models through services like Amazon Bedrock and Google Vertex AI, allowing customers to run Claude or Gemini within Ona environments. Under OpenAI ownership, this multi-model flexibility may be compromised. The deal also highlights fundamental challenges in autonomous AI deployment. While customer-controlled execution addresses where agents operate, it doesn't solve the problem of ensuring agent accuracy during extended, unattended operations. Tools for reviewing and validating long-running autonomous tasks remain underdeveloped compared to the execution environments that host them. For enterprise decision-makers, this acquisition represents both opportunity and risk. Organizations gain access to powerful AI capabilities while maintaining security control, but they also assume operational responsibility for the infrastructure supporting these agents. The success of this approach will depend on OpenAI's commitment to maintaining Ona's enterprise-friendly features while integrating them effectively with Codex capabilities.

OpenAI's acquisition of Ona marks a significant shift in the enterprise AI landscape, addressing one of the most pressing concerns preventing widespread adoption of autonomous coding agents in regulated industries. The German startup, which evolved from developer tools company Gitpod, brings critical enterprise security and governance capabilities to OpenAI's Codex platform.

The acquisition announcement reveals the scale of Codex's growth and the challenges that come with it. OpenAI reports that Codex now serves more than 5 million users weekly, representing a 400% increase since the beginning of 2026. More significantly, the nature of work performed by these agents has evolved from quick, minute-long tasks to complex projects spanning hours or even days. This extended operation timeline creates new requirements for persistent execution environments and raises critical questions about enterprise security and governance.

Ona's value proposition centers on solving the trust equation for enterprise AI deployment. The company's platform architecture addresses three fundamental enterprise requirements: secure execution environments, autonomous agent capabilities, and comprehensive governance controls. Their sandboxed cloud workspaces provide consistent, code-defined deployment environments, while background agents can accept complex tasks and deliver results through pull requests accessible from any device.

The security and governance layer represents Ona's most valuable contribution to the OpenAI ecosystem. Enterprise customers gain access to detailed audit trails, role-based access controls, scoped credentials, and the ability to deploy agents within their own virtual private clouds. This "customer-controlled execution" model allows organizations to leverage OpenAI's AI capabilities while maintaining complete control over their data, credentials, and security policies.

The competitive implications of this acquisition are substantial. OpenAI faces direct competition from Anthropic, whose Claude Code has gained significant momentum among engineering teams over the past year. Both companies have filed confidential SEC prospectuses, indicating preparation for potential public offerings. With OpenAI's valuation reaching approximately $852 billion, the pressure to convert AI enthusiasm into sustainable enterprise revenue has intensified.

This acquisition fits within OpenAI's broader enterprise strategy, following previous purchases of security testing startup Promptfoo and healthcare technology firm Torch. These deals prioritize enterprise infrastructure and compliance capabilities over core AI model improvements, demonstrating OpenAI's recognition that enterprise adoption requires more than advanced AI capabilities.

However, the acquisition raises important concerns about market competition and customer choice. As an independent vendor, Ona supported multiple AI models through integration with services like Amazon Bedrock and Google Vertex AI. Customers could deploy Claude, Gemini, or other models within Ona's secure environments. Under OpenAI ownership, this model neutrality may be compromised, potentially forcing customers into vendor lock-in situations.

The deal also highlights fundamental limitations in current autonomous AI technology. While customer-controlled execution addresses where agents operate, it doesn't solve the accuracy and reliability challenges inherent in long-running autonomous tasks. An agent that operates incorrectly for two days can cause significant damage, and the tools for monitoring, reviewing, and correcting extended autonomous operations remain immature compared to the execution environments that host them.

For enterprise decision-makers evaluating this development, several critical questions emerge. Will Ona maintain its model-neutral approach under OpenAI ownership, or will it gradually favor Codex over competing solutions? Which security and governance features represent genuine Ona innovations versus standard cloud provider controls that enterprises already purchase? How will review and rollback mechanisms function when agents have been operating within production systems for extended periods?

The acquisition's success will ultimately depend on regulatory approval and OpenAI's ability to maintain Ona's enterprise-friendly characteristics while integrating them with Codex. The deal faces potential scrutiny from regulators on both sides of the Atlantic, given the implications of a US technology giant acquiring German infrastructure technology.

This development signals a broader shift in the AI industry, where competition is moving beyond model capabilities to focus on execution infrastructure and enterprise trust. Organizations that previously hesitated to deploy autonomous agents due to security and governance concerns now have a potential path forward, assuming OpenAI maintains Ona's commitment to customer-controlled execution and multi-model flexibility.

Посилання в статті