Over the past several weeks, I've written a series of articles examining the risks associated with artificial intelligence. I've explored how AI agents can behave contrary to their developers' intentions, how AI systems can be attacked, and what MITRE's ATLAS catalog tells us about those attacks.
The articles required considerable research, including analysis of hundreds of attack techniques and dozens of documented case studies. They've also required something less easily measured: my judgment. I've had to decide which evidence matters, how to interpret it, and what conclusions it reasonably supports.
Not everyone will agree with those conclusions. In fact, a reader recently challenged some of my interpretations of the ATLAS data, raising questions about terminology, evidence, and the standards to which FAIR Institute publications should be held.
I disagreed with parts of that criticism, but I also found some of it persuasive. More importantly, it prompted me to consider a broader question: What role should opinion articles play in an organization whose mission includes developing authoritative professional standards? I believe now is a time to ask this question as we transform how the FAIR Institute operates.
The FAIR Institute exists to advance the discipline of technology risk management, with an emphasis on cybersecurity- and AI-related concerns. An important part of that mission is developing standards that establish a common language, consistent models and methods, and solid program principles that organizations rely upon.
As you'll learn at FAIR Forward 2026, we're preparing to significantly expand our standards development efforts, with greater community participation and a stronger emphasis on establishing authoritative guidance across the cyber and AI risk management disciplines. That makes it increasingly important to distinguish between the Institute's formal standards and the opinions expressed by individual contributors, including those of us in Institute leadership like myself.
Our standards are developed through structured processes and chartered working groups involving subject matter experts and community review. Once approved by our standards committee, they represent the Institute's authoritative guidance. Our blog posts and other less formal content channels serve a different purpose. They provide a forum for exploring emerging issues, interpreting data, questioning established thinking, and proposing ideas that may not yet have achieved consensus.
These activities aren't competing priorities. In fact, they're mutually reinforcing.
Standards development shouldn't happen in an intellectual vacuum. Before we can establish consensus around an approach, we need opportunities to examine alternatives, challenge assumptions, and learn from others in our community who see things differently. Blogs, social media posts, working groups, community discussions, and other forums can help us do that.
This is particularly important in rapidly evolving areas such as AI risk management, where both the technology and its associated risks are changing faster than formal standards can reasonably evolve. Waiting for consensus before expressing an opinion would mean remaining silent precisely when the profession needs informed discussion.
Of course, the freedom to express opinions doesn't excuse careless research or unsupported conclusions. Nor should readers have to guess whether something represents an author's perspective or the Institute's official position. We haven't always made that distinction sufficiently clear, and we intend to improve that, beginning with a standard disclaimer on recent opinion articles, including those I've recently published.
The objective isn't to make our blogs less authoritative by labeling them as opinions. It's to make clear where their authority comes from: the evidence, reasoning, and expertise behind the argument, rather than the institutional position of the author.
Let me return to the criticism of my blog post from last week involving MITRE ATLAS.
Among other things, the reader questioned my characterization of MITRE's mitigations as controls and whether the absence of mitigation mappings justified a section titled, "The Controls Are Lagging." Per MITRE ATLAS, mitigations "describe security concepts, practices, and technologies that can reduce the likelihood or impact of adversary Techniques and sub-Techniques." MITRE maps its mitigations to the attack techniques that they address.
I stand behind the use of the term controls, which is consistent with the broad definition in the FAIR Controls Analytics Model (FAIR-CAM) standard. There, controls are defined as "anything that can be used to reduce the frequency or magnitude of loss." MITRE's definition is logically consistent with FAIR-CAM's broad definition of controls, because likelihood (e.g., 30% chance of happening this year) is calculated directly from frequency (e.g., this happens once every 3 years) and impact generally describes the nature and magnitude of loss. I could have made this point in my article to remind readers of the FAIR-CAM definition and demonstrate why I believe "controls" is the appropriate term, but I did not.
But our reader's second criticism reveals another lesson for me. The data showed that ATLAS had fewer mitigation mappings than MITRE ATT&CK, meaning that fewer adversary techniques in the ATLAS catalog are covered by mitigations than those in the ATT&CK catalog. It didn't establish that effective controls weren't available at all for the unmapped techniques, just that MITRE had not identified or mapped them. I had acknowledged that uncertainty in the article itself, but the heading may have appeared more "sky is falling" than "there's work to do." I should be more careful in my choice of headlines.
The reader also suggested that the article should have explained how ATLAS data could inform FAIR risk analysis. That's certainly an important question, but it wasn't the purpose of that particular article. The preceding post introduced the AI attack surface, and this ATLAS article explored how attackers exploit the different components of that surface. I am writing a subsequent post that will examine how that information can inform a quantitative risk analysis.
The point isn't to adjudicate every disagreement here. It's that these exchanges are valuable because reasonable professionals can interpret the same information differently. (Sounds familiar, right?) And the reader was concerned specifically about the use of standard terminology, advocating that we must be vigilant in how we use it in our blog posts and other content. I agree.
Here's where I have some concerns: the reader suggested we put a light peer review process in place for blog posts. I don't know if the delay in publishing a blog post that results from a peer review process, even a "light" one, would be justified in all cases. I think it should be case-by-case, depending on the subject, content, author, and other factors.
There's a risk with this approach. Choosing to publish a post before it has been subjected to a review increases the possibility of errors, unsupported inferences, and conclusions that later prove mistaken. Choosing to delay one to allow for peer review stifles the free flow of ideas and places a burden on scarce subject matter experts. Delays often frustrate authors. Indeed, I've held up a blog post or two submitted by our members (Laura and Caleb, I'm thinking of you!) because I didn't have the spare cycles to review it in a timely manner.
I believe there's room for a discerning approach: we expect authors to perform the necessary research, be transparent about their reasoning, and remain open to informed criticism, without requiring every blog post to go through a review.
As the FAIR Institute expands its standards development efforts, we'll provide more opportunities for members to contribute ideas, challenge assumptions, and participate in developing authoritative guidance. These may come through formal working groups and our standards committee, but to get the broadest exposure of and reaction to new and evolving ideas, we must use blogs, social media, webinars, conferences, and other less formal mechanisms without overly burdensome processes that hinder discussion and debate.
The distinction between those activities matters. A published standard should carry the weight of the process through which it was developed and approved. An opinion article should stand on the strength of its evidence and reasoning, regardless of who wrote it. But the latter can and should support the development of the former.
Today's controversial idea may become tomorrow's accepted practice or even a FAIR Institute standard. Equally, an idea that initially seems compelling may prove mistaken when subjected to scrutiny. Both outcomes represent progress if we're willing to follow the evidence.
That's why I welcome thoughtful criticism of my articles, even when I disagree with it. I don't expect every conclusion I publish to survive scrutiny unchanged, particularly when exploring subjects as new and complex as AI risk. The process of researching, writing, questioning, and debating those conclusions will leave us with a better understanding than we had before.
The FAIR Institute's growing commitment to standards development should strengthen this process, not constrain it. We need authoritative guidance that professionals can trust, and we need forums where ideas can be explored before they're ready to become authoritative.
I believe a healthy professional community makes room for both. Its credibility depends not on always being right, but on being rigorous enough to recognize when it isn't, and willing to learn from the difference.
Do you agree? If not, please keep your criticism to yourself. (Just kidding. I think.)