The last two weeks gave the model act remedies that bite — suppression and civil damages. But I ended both posts on the same admission, and it is time to deal with it. Suppression and damages are reactive. They need a victim who knows. The defendant files a motion; the wronged person files a suit. Neither does anything for the far larger group whose data was over-collected, over-retained, queried, or repurposed and who never find out it happened. For them, the two remedies I spent two weeks building are a light switch in an empty room. So the model act needs one more kind of protection — the kind that does not wait for anyone to complain. It needs sunlight.
The principle is simple: you cannot deter what no one can see. Reporting does not remedy any one person’s harm. It does something the individual remedies cannot — it makes the aggregate visible even when every individual case stays hidden. It lets Congress, the courts, the press, and the public see the shape of what the government is doing with digital property and react to a trend before a scandal forces the issue. Counting is itself a discipline. An agency that must publish how much it collects, queries, and keeps is an agency that must know those numbers — and knowing them changes behavior even before anyone outside reads them.
Title III, as usual, wrote the template. Section 2519 requires an annual Wiretap Report: every year, judges and prosecutors report each intercept order — how many were authorized, how many extended, how long they ran, what they cost, how many communications were intercepted, how many were incriminating, and how many arrests and convictions followed. The Administrative Office of the U.S. Courts compiles it and publishes it. It is public, numeric, and annual. It is not perfect, but it exists, and it has produced a decades-long time series that lets anyone watch the trend lines move. That is the model — and, as with the rest of Title III, the model needs serious updating before it fits the digital world.
Here is the flaw that matters most. Section 2519 counts the front door. It reports authorizations — the government asking permission to collect. It reports almost nothing downstream. It does not count how many times the government later queried or re-searched data it already holds, how much it retained past the purpose that justified collection, how often it used that data for something new, how many disclosures it made, or — now — whether it fed any of that content into a model. In a world where the real intrusion happens at analysis, not collection — where, as I keep arguing, the query is the search — a report that counts only the authorizations misses the surveillance that actually bites. FISA is the cautionary tale: the statistical transparency reports that eventually surfaced hundreds of thousands of non-compliant queries came out years late and in the aggregate, long after the practice had entrenched. Late transparency about the front door is not transparency about the house.
So § 11 does five things, and the first is to move the count downstream.
First, report the whole lifecycle, not just the authorization. The annual report should count queries and analyses run against retained data, the volume of data retained and for how long, secondary uses of collected content — including any use as training data for a model — disclosures made, and dispositions carried out: what was purged, and when. Every one of these is a place the earlier sections drew a line, and a line no one counts is a line no one keeps.
Second, report the denominator, not just the numerator. Section 2519 tells you how many intercepts were authorized. It does not tell you how many people’s communications were swept in per authorization, or how many of those people were ever charged. That ratio — how many were collected against how many were accused — is the single most honest measure of over-collection, and it is exactly the number no current report requires. Publish it, and the scale of the bystander problem stops being an abstraction.
Third, build in independent oversight with access to the handling record. Section 2519 is self-reporting — the government counting itself. That is necessary but not sufficient. Section 11 should pair the public report with an independent auditor or inspector general who has access to the chain-of-custody-for-data that § 6 requires, empowered to test the self-reported numbers against the actual logs and to publish findings. Self-reporting tells you what the agency chose to say; an auditor with the records tells you what the records show. The handling record is what makes the difference between a report you have to trust and a report you can check.
Fourth, make it timely and granular. Annual, promptly published, and broken down by authority, agency, data type, and purpose — aggregated and anonymized to protect individuals, but granular enough to be meaningful. A single national total hides everything; a breakdown lets a trend in one program or one data type surface. And current matters: the FISA lesson is that a number three years late is a number that arrives after the fight it should have informed is over.
Fifth — and this is this week’s direct payoff — reporting is what would drag the training-data problem into the light. Last Thursday I argued that the government using seized digital property to train AI models is the use with no undo, and that its defining feature is invisibility: you never learn your data is in a model. A lifecycle reporting duty that requires the government to disclose which categories of collected data were used to train or fine-tune which models turns that invisible use into a countable, auditable fact. Transparency is what makes the § 10 training-data remedy enforceable in the first place — you cannot sue over a use you cannot discover, and reporting is a second path, alongside individual notice, for the use to be discovered at all.
Now the objections, and they are the serious ones agencies always raise. The first is operational security: detailed reporting tips off adversaries to sources and methods. But aggregate statistics are not methods. Section 2519 has published wiretap counts for decades without compromising a single wire; the genuinely sensitive slivers can go in a classified annex while the public still gets the numbers. “We can’t tell you how much we collect without helping the enemy” is almost never true at the level of a yearly total, and where it is, the annex handles it.
The second objection is burden: counting all of this is expensive. But if an agency cannot say how much data it is collecting, querying, and keeping, that is not a reason to excuse it from reporting — it is evidence the agency is holding more than it can account for, which is the problem. And the handling record under § 6 already generates the underlying data. Reporting does not create a new bureaucracy; it surfaces what a well-run system already knows about itself.
The third objection is that reporting is toothless — numbers in a document no one reads. Sometimes. But numbers are the raw material of every other form of accountability: they feed litigation, journalism, and the reauthorization fights where surveillance authorities actually get reined in. The FISA query numbers, late as they were, drove real reform pressure once they existed. And pairing the report with an independent auditor gives it the teeth that self-reporting alone lacks. A number someone is empowered to verify and publish is not toothless.
Step back and here is why I call reporting the keystone rather than the appendix. The first four sections govern what the government may do. The remedy sections react when it does wrong in a visible case. But reporting is the only mechanism that operates continuously, on everything, whether or not anyone complains — which means it is how the other sections get enforced in the dark, in the vast majority of cases that never produce a motion or a suit. A surveillance statute without a real reporting section is a lock that no one ever checks is being used. The sunlight is not decoration. It is the enforcement.Next Tuesday closes Pass 1 with § 12: provider assistance — the private companies the government leans on to do the collecting in the first place, and the rules that should govern what we conscript them to do.
Leave a Reply