A board paper about AI safety usually does one job in its opening pages, which is to establish that the threat surface is moving faster than the controls designed for it. The work the paper is asking for — a new capability, a standing committee, a mandated control set — exists because that gap exists. Everything in the body inherits from that opening. Get the urgency wrong and the Ask reads as a wish list. Get it right and the Ask reads as the only reasonable answer to a situation the reader now feels.
This is why the question of what counts as a citation in those opening pages is load-bearing in a way that doesn’t feel obvious until you watch it go wrong. The instinct, when something AI-shaped lands in the news during a drafting cycle, is to reach for it. It is current. It is on-topic. Adding it feels like keeping the paper alive. But the test isn’t whether a piece of news is topically adjacent to what the paper is about. The test is whether it sits in the same source class as the rest of the urgency anchors — and whether its rhetorical effect points the same way as the case being built.
Take the OpenAI Daybreak launch from the second week of May 2026, a frontier-AI product for cyber defenders bundling models and code generation and security-partner distribution into a continuous defensive loop. Now imagine a paper whose opening pages anchor urgency in the canonical sources of the moment — government-commissioned scientific synthesis on AI capability trends, evaluations from a national safety institute on a recently released frontier model, lab-side disclosures about withholding a successor model from public release on capability-risk grounds, and regulator and central-bank statements escalating the supervisory weight on AI monitoring. Daybreak is topically adjacent to all of it. It is also, as evidence, the wrong shape.
The shape is wrong in two ways at once. The first is source class. Those anchor citations are regulator, government-commissioned synthesis, and lab-side capability-risk disclosure. They carry weight because they sit above the vendor surface. A product launch tweet, however well-engineered the product behind it, is marketing. The asymmetry between the anchor sources and the candidate addition is visible to any reader paying attention. Including the addition lowers the average altitude of the paper’s evidence base and, worse, signals that the drafters are reaching.
The second is rhetorical direction. The threat-surface case wants the reader to feel a widening gap — capabilities going up faster than the controls catching them. Daybreak, in that frame, points the other way. Its subtext is that vendor mitigation is already arriving, that the gap is closing on its own, that the threat-surface is being met. The literal content of the product launch is almost beside the point; the act of citing it inside a paper arguing the opposite case carries a counter-current. The reader doesn’t have to articulate the dissonance to feel it. They just close the page slightly less convinced than they opened it.
The general rule under both observations is the same. When the news cycle hands you an artefact during a drafting window, the question to ask is not whether it relates to the topic. The question is whether it sits at the source-class altitude the paper has already set and whether its rhetorical vector reinforces or opposes the case the paper is making. Topical adjacency tells you almost nothing. Source-class fit and direction-of-effect tell you most of what matters. If the answer to either is wrong, the news is not a citation. It is a capture target for another file, and the paper is better for not having it.