Abnormal joins OpenAI's Daybreak cyber partner programme

Abnormal AI has joined OpenAI's Daybreak Cyber Partner Program, expanding ties between the two companies in enterprise AI and cyber defence.

Under the partnership, Abnormal will broaden its use of OpenAI tools across its operations and product development while the companies explore new security features for customers. They will examine how OpenAI's frontier models and Codex Security could work with Abnormal's Behavioural AI to help security teams monitor AI agent activity and detect anomalous behaviour.

The agreement connects two companies with different positions in the market. Abnormal focuses on email, identity and insider threat protection, using behavioural models to detect attacks it says evade conventional rules-based systems. OpenAI has been building a cyber security offering that includes models for threat investigation, vulnerability validation and software remediation.

For Abnormal, the partnership also points to broader internal use of generative AI. It will expand its use of OpenAI across business operations and software development, while also working with OpenAI on practical approaches enterprises can use to roll out AI in a controlled way.

That work is expected to draw on lessons from engineering, customer operations and security workflows. The collaboration centres on governance and oversight as companies deploy more AI systems into daily work, particularly autonomous or semi-autonomous agents that may act across multiple business applications.

Security focus

OpenAI's Daybreak Cyber Partner Program brings together security vendors and service providers to develop tools for defenders. Here, the immediate focus is monitoring AI-driven activity and identifying behaviour that falls outside expected patterns.

This is becoming a more pressing issue for security teams as businesses adopt AI agents for coding, analysis and administrative tasks. As these systems spread, organisations need to understand what agents are doing, whether their actions align with policy, and whether unusual behaviour signals misuse, compromise or error.

Abnormal said its core detection and response products will continue to rely on its own Behavioural AI. That suggests the partnership is meant to complement, not replace, the company's existing technology in areas where OpenAI's models may add investigative or workflow support.

Abnormal says it protects more than 4,500 organisations, including more than a quarter of the Fortune 500. That installed base gives it a large audience for any new security tools that emerge from the partnership, particularly among large companies trying to balance AI adoption with tighter controls.

OpenAI has been seeking partnerships that place its cyber products into established security workflows rather than requiring customers to adopt entirely new operating models. The strategy relies on working with vendors that already have customer relationships and domain expertise in specific areas of cyber security.

Abnormal's role in email and identity security fits that approach. These are areas where malicious activity often first appears, whether through phishing, account takeover, business email compromise or suspicious access patterns involving internal users and third parties.

"We partner with companies that are aligned to our mission of fighting crime, and that want to bring customers into the future with the most effective AI," said Evan Reiser, Founder and Chief Executive Officer of Abnormal AI.

"That is why Abnormal is expanding its use of OpenAI for internal business operations and software development, and why we are working together to help every enterprise adopt AI with confidence," said Reiser.

Broader push

The deal also reflects a wider shift in cyber security, as suppliers try to apply large language models and agent-based systems to both defensive operations and internal software work. Vendors argue that these tools can help analysts investigate incidents faster, review code, validate vulnerabilities and reduce repetitive work, but customers remain cautious about control, explainability and data handling.

OpenAI's Daybreak effort is aimed at that tension. The programme combines its cyber models, Codex Security and controlled access arrangements with partners that can embed those tools in products already used by defenders.

For enterprises, one of the central questions is whether AI can be introduced without weakening oversight. That is especially relevant when AI agents may take actions across security, engineering or operational systems with limited human intervention.

Mike Aiello, Head of Cyber Products at OpenAI, described the company's view of Abnormal's position in the market.

"Abnormal brings deep expertise in email and identity security, and its products are already part of many workflows enterprises use to protect their people and systems," said Mike Aiello, Head of Cyber Products at OpenAI.

"By partnering, we will help accelerate the adoption of practical defenses and bring advanced AI capabilities to more defenders through tools they already trust," said Aiello.