OpenAI's Security Acquisition, Hardware Exit, and the Coding Debate

Three distinct threads ran through OpenAI’s week: a defensive security bet on agentic AI, a high-profile leadership departure tied to its military pivot, and a chairman’s candid take on what coding looks like in an AI-first era.

Key points

• OpenAI plans to acquire Promptfoo, an AI security platform focused on agentic testing and evaluation.
• Hardware chief Caitlin Kalinowski resigned following OpenAI’s Pentagon deal, citing concerns about surveillance and lethal autonomy.
• Chairman Bret Taylor commented publicly on the future of writing code by hand in an AI-assisted world.
• The Promptfoo deal signals OpenAI is treating security as a core infrastructure problem as its agent products scale.
• Kalinowski’s exit adds to a pattern of senior departures tied to OpenAI’s expanding defense relationships.

Why is OpenAI buying a security firm now?

Promptfoo is an open-source AI security platform built around testing and evaluating AI systems for vulnerabilities. OpenAI’s reported acquisition is aimed at accelerating agentic security testing capabilities, which is a telling priority given where the product roadmap is heading.

Agentic AI systems, ones that take sequences of real-world actions with limited human oversight, introduce a different class of risk than standard chatbots. An agent that can browse the web, write and execute code, or interact with external APIs is also an agent that can be manipulated, misdirected, or exploited in novel ways. Existing security frameworks were not built with these threat models in mind.

Bringing Promptfoo in-house suggests OpenAI wants this capability embedded in its engineering process rather than bolted on. For enterprise and government customers, particularly those evaluating OpenAI’s operator platform, security evaluation tooling is increasingly a procurement requirement. The acquisition, if completed, would give OpenAI something concrete to point to. Terms were not disclosed in available reporting.

What does Kalinowski’s resignation signal?

Caitlin Kalinowski had been OpenAI’s head of hardware. Her departure, reported March 8, came in the wake of OpenAI’s deal with the Pentagon and was tied to her concerns about surveillance applications and lethal autonomy in AI systems.

This is a materially different kind of exit from the departures OpenAI has seen before. Kalinowski did not leave for a competitor or to start a new venture, at least not as the primary stated reason. The concerns she raised touch on a genuine tension inside the company. OpenAI revised its usage policies in early 2024 to allow certain national security applications, and its relationship with defense customers has deepened since. Those decisions have not been universally welcomed internally.

For investors tracking OpenAI’s path to commercial maturity, the hardware role itself deserves attention. OpenAI has signaled ambitions in custom silicon and physical AI infrastructure, and a leadership gap there carries operational weight beyond the reputational optics of the exit.

What did Bret Taylor actually say about coding?

Taylor’s comments, reported by Business Insider, touched on the practice of writing code by hand in an environment where AI coding tools are increasingly capable. The source summary is truncated, so the full context of his remarks is not available from the provided reporting.

What is clear is that the OpenAI chairman is engaging publicly with a question that matters to a significant portion of OpenAI’s target market. Developers are both the primary users of tools like Codex and GitHub Copilot (which competes with OpenAI’s own coding products) and a constituency with strong opinions about AI-assisted workflows. Taylor speaking on this topic is consistent with OpenAI’s push to own the developer-tools conversation, a thread this publication noted in the February 22 edition covering OpenAI’s coding ambitions.

The remarks carry more weight given Taylor’s background. He co-founded Quip, served as Salesforce co-CEO, and has deep product credibility with technical audiences. His framing of the hand-coding question, whatever it was precisely, will land differently than if a policy executive had said it.

The broader picture this week

Taken together, the three stories this week sketch a company managing growth in several directions at once. The Promptfoo acquisition is proactive, a bet that security infrastructure will be a competitive differentiator as agents proliferate. The Kalinowski departure is reactive, a consequence of strategic choices that not everyone inside the company agrees with. And Taylor’s public commentary is positional, aimed at keeping OpenAI credible with developers as it competes aggressively in the coding-tools market.

None of these stories changes the fundamental commercial trajectory. But the Kalinowski exit in particular is worth watching. If OpenAI’s defense partnerships continue to expand, internal friction over those relationships could affect hiring and retention in technical and research roles, which is where the company’s long-term product edge actually lives.