The national intelligence community (NIC) has spent the past quarter-century building effective collection, processing and analytic capability. But if intelligence arrives in the wrong form, at the wrong speed or in a way poorly matched to how consumers (think ministers, officials and operators) increasingly absorb information, it will lose influence. That is the central argument of my new ASPI report, Reading the Room: Redesigning intelligence product for the AI age. In the emerging AI era, the looming issue is whether intelligence agencies can deliver their hard-won insights in a form decision-makers can and will actually use.
That matters because AI is already changing expectations about information throughout society and government. Generative AI is training everyone to expect conversational search, rapid synthesis, tailored and personalised outputs, and answers delivered in formats suited to the task at hand. Those habits will not stop at the door of national security decision-making. Intelligence consumers will increasingly expect classified systems to be as responsive, accessible and interactive as non-classified systems. If the NIC doesn’t adapt carefully and effectively, it risks being outpaced not just by adversaries, but by the expectations of its own customers.
This is not a cosmetic problem. It goes to how to generate and sustain real national intelligence power. Ministers operate under intense time pressure. Senior officials are inundated with competing inputs. Operators need relevance now, not elegant prose after the decision window has closed. Yet much intelligence product still assumes a world in which the report itself is the endpoint. It is not. In an AI-shaped information environment, the traditional intelligence report is increasingly just one interface between insight and decision.
In the report I sketch three all-too plausible (and for consumers, increasingly desirable) scenarios to show where this could lead: a secure AI-enabled interface that lets customers query intelligence holdings directly, thereby generating new, potentially unmediated intelligence judgements; intelligence products hyper-personalised to the expressed and implicit wishes of different consumers, in both format and content; and automated sanitisation and dissemination that gets intelligence to operators, departments and partners more quickly. None of this is especially exotic. These are tasks humans already can and do perform – just not at the extraordinary scale, pace and cost promised by AI.
The upside is obvious. Using AI in these ways could help close the distance between insight and decision. It could make intelligence more timely, more accessible and more relevant to the context in which decisions are actually made. For Australia, that is not a nice-to-have. We face a tougher strategic environment, tighter warning times and adversaries who will themselves exploit AI for speed, scale and disruption wherever they can. Intelligence that cannot keep pace with that reality will matter less precisely when it needs to matter most.
But this is not an argument for technological boosterism. AI does not understand in the human sense. It predicts, imitates and generates plausible outputs. That creates real risks for intelligence product: subtle and not-so subtle errors, automation bias, false confidence, opaque accountability and systems that become easier to steer towards the answers consumers want to hear rather than the answers they need. And what if the price of achieving effect through hyper-personalisation is that shared understanding and effective decision-making is undermined?
Indeed, all the upsides identified in the scenarios have their mirror-image downsides that can’t be ignored. So, the case for adaptation is not a case for surrendering judgement to machines; it is a case for redesigning intelligence product, including in ways that make human judgement more visible, not less.
That means the NIC should start treating intelligence consumption as a first-order strategic issue. It should learn from allies and from adjacent knowledge industries already grappling with AI-driven change. It should prepare its workforce not only to use AI, but also to explain it, challenge it and govern it. And it should redesign its product rather than simply digitise legacy forms.
Australia also needs to think about this in consciously sovereign terms. We should learn from the United States and other Five Eyes partners, but we should not outsource the conceptual work to them. AI will almost certainly widen existing asymmetries between large and middle powers. If Australia does not decide for itself how its intelligence products should be designed, trusted and governed in this environment, others will make those decisions by default.
That is why this debate should not be left to technologists alone. The NIC cannot meet the AI age with an analogue product and assume enduring relevance. It needs to adapt with urgency, discipline and a clear sense of what makes intelligence distinctive in the first place: judgement, credibility, contestability and service to the national interest. In the decade ahead, the real test will not simply be whether agencies know more; it will be whether they can turn knowledge into real decision advantage before someone else does.
