One of the more encouraging changes in classification policy over the past decade has been the sharp reduction in the number of decisions to classify information reported each year by executive branch agencies.
In 2005 there were a total of 258,633 original classification actions, or new secrets, reported; in 2015, there were said to be 53,425 such actions. (See Number of New Secrets in 2015 Near Historic Low, Secrecy News, July 29, 2016).
Despite the misleading precision with which they are reported, these numbers — which are derived from agency reports to the Information Security Oversight Office and published in ISOO annual reports — were understood to be estimates, not precise tabulations.
Now, however, a new report from the State Department Inspector General suggests that State’s reporting of its classification activity to ISOO may not only be imprecise, but actually inaccurate and incorrect.
The Inspector General “found shortcomings with the count of classification decisions” reported to ISOO. The estimates that were generated were not validated, and they did not reflect the full scope of State Department classification activity.
So, “For example, classified documents created within the Office of the Secretary were not included” in the survey, the IG said. See Compliance Follow-up Review of the Department of State’s Implementation of Executive Order 13526, Classified National Security Information, Office of Inspector General, Department of State, September 2016.
The bottom line, the IG said, is that reported classification totals “will not accurately represent all of the Department’s classification decisions because not all decisions are being identified or sampled as part of the Department’s self-inspection program.”
William Cira, the acting director of the Information Security Oversight Office, said he was not surprised by the Inspector General findings, and not especially troubled.
He recalled that ISOO itself stated in its 2009 report that “the data reported has not truly reflected the changing ways agencies have generated and used classified information in the electronic environment.”
“It has been recognized, even long before we asked the agencies to include the electronic environment, that an actual count is not feasible,” Mr. Cira added. “The sampling and extrapolation technique described in that report has been in widespread use for a long time.”
“It is actually one of the suggested methods that we impart to the agencies when we send out our data collection request each year. Since FY 2009, ISOO has asked agencies to do their best to estimate the volume of all classified products in the electronic environment.”
“We have always acknowledged that this would not be easy. We do ask them [agencies] to estimate, we do suggest that they sample and extrapolate, and we acknowledge that in almost all cases they will not have the resources to conduct a scientific survey as that is defined by professional statisticians.”
“This method may seem crude but we recognize that almost none of agency data collectors have trained statisticians to call upon, and there is no expectation that they hire one.” Still, “If the Dept. of State OIG believes that the Office of the Secretary should be included that is a welcome suggestion.”
“The one thing for certain is that this method has been consistently applied across many agencies for a very long time,” Mr. Cira said.
In other words, if the collection method is crude, at least it is consistent in its crudeness, and so perhaps some rough trend information may still be discerned within the noise.
But without real quantitative and qualitative clarity, effective management of agency classification activity will be beyond reach.
These ideas aim to advance the detailed policy solutions needed to foster public trust and implement fairness in the adoption of AI across diverse domains, from healthcare and government benefits to rural access, education, and worker protections.
The evidence is clear: algorithmic pay-setting is established in app-based work, and payroll/timekeeping failures show how software can produce systemic wage harm at scale
While a few states have taken steps to implement decision-making mechanisms for certain AI systems, too many leaders are simply accepting narratives about AI’s purported public benefit at face value – jumping to the “how” of AI implementation before thoroughly vetting potential systems and deciding whether they are appropriate to use at all.
When properly structured — with specific numeric targets, secured financial obligations, independent monitoring, and meaningful enforcement — CBAs transform data center deals into durable community partnerships.