As I noted in my argument that CISA is designed to do what NSA and FBI wanted an upstream cybersecurity certificate to do, but couldn’t get FISA to approve, there’s almost no independent oversight of the new scheme. There are just IG reports — mostly assessing the efficacy of the information sharing and the protection of classified information shared with the private sector — and a PCLOB review. As I noted, history shows that even when both are well-intentioned and diligent, that doesn’t ensure they can demand fixes to abuses.
So I’m interested in what Richard Burr and Dianne Feinstein did with Jon Tester’s attempt to improve the oversight mandated in the bill.
The bill mandates three different kinds of biennial reports on the program: detailed IG Reports from all agencies to Congress, which will be unclassified with a classified appendix, a less detailed PCLOB report that will be unclassified with a classified appendix, and a less detailed unclassified IG summary of the first two. Note, this scheme already means that House members will have to go out of their way and ask nicely to get the classified appendices, because those are routinely shared only with the Intelligence Committee.
Tester had proposed adding a series of transparency measures to the first, more detailed IG Reports to obtain more information about the program. Last week, Burr and DiFi rolled some transparency procedures loosely resembling Tester’s into the Manager’s amendment — adding transparency to the base bill, but ensuring Tester’s stronger measures could not get a vote. I’ve placed the three versions of transparency provisions below, with italicized annotations, to show the original language, Tester’s proposed changes, and what Burr and DiFi adopted instead.
Comparing them reveals Burr and DiFi’s priorities — and what they want to hide about the implementation of the bill, even from Congress.
Prevent Congress from learning how often CISA data is used for law enforcement
Tester proposed a measure that would require reporting on how often CISA data gets used for law enforcement. There were two important aspects to his proposal: it required reporting not just on how often CISA data was used to prosecute someone, but also how often it was used to investigate them. That would require FBI to track lead sourcing in a way they currently refuse to. It would also create a record of investigative source that — in the unlikely even that a defendant actually got a judge to support demands for discovery on such things — would make it very difficult to use parallel construction to hide CISA sourced data.
In addition, Tester would have required some granularity to the reporting, splitting out fraud, espionage, and trade secrets from terrorism (see clauses VII and VIII). Effectively, this would have required FBI to report how often it uses data obtained pursuant to an anti-hacking law to prosecute crimes that involve the Internet that aren’t hacking; it would have required some measure of how much this is really about bypassing Title III warrant requirements.
Burr and DiFi replaced that with a count of how many prosecutions derived from CISA data. Not only does this not distinguish between hacking crimes (what this bill is supposed to be about) and crimes that use the Internet (what it is probably about), but it also would invite FBI to simply disappear this number, from both Congress and defendants, by using parallel construction to hide the CISA source of this data.
Prevent Congress from learning how often CISA sharing falls short of the current NSA minimization standard
Tester also asked for reporting (see clause V) on how often personal information or information identifying a specific person was shared when it was not “necessary to describe or mitigate a cybersecurity threat or security vulnerability.” The “necessary to describe or mitigate” is quite close to the standard NSA currently has to meet before it can share US person identities (the NSA can share that data if it’s necessary to understand the intelligence; though Tester’s amendment would apply to all people, not just US persons).
But Tester’s standard is different than the standard of sharing adopted by CISA. CISA only requires agencies to strip personal data if the agency if it is “not directly related to a cybersecurity threat.” Of course, any data collected with a cybersecurity threat — even victim data, including the data a hacker was trying to steal — is “related to” that threat.
Burr and DiFi changed Tester’s amendment by first adopting a form of a Wyden amendment requiring notice to people whose data got shared in ways not permitted by the bill (which implicitly adopts that “related to” standard), and then requiring reporting on how many people got notices, which will only come if the government affirmatively learns that a notice went out that such data wasn’t related but got shared anyway. Those notices are almost never going to happen. So the number will be close to zero, instead of the probably 10s of thousands, at least, that would have shown under Tester’s measure.
So in adopting this change, Burr and DiFi are hiding the fact that under CISA, US person data will get shared far more promiscuously than it would under the current NSA regime.
Prevent Congress from learning how well the privacy strips — at both private sector and government — are working
Tester also would have required the government to report how much person data got stripped by DHS (see clause IV). This would have measured how often private companies were handing over data that had personal data that probably should have been stripped. Combined with Tester’s proposed measure of how often data gets shared that’s not necessary to understanding the indicator, it would have shown at each stage of the data sharing how much personal data was getting shared.
Burr and DiFi stripped that entirely.
Prevent Congress from learning how often “defensive measures” cause damage
Tester would also have required reporting on how often defensive measures (the bill’s euphemism for countermeasures) cause known harm (see clause VI). This would have alerted Congress if one of the foreseeable harms from this bill — that “defensive measures” will cause damage to the Internet infrastructure or other companies — had taken place.
Burr and DiFi stripped that really critical measure.
Prevent Congress from learning whether companies are bypassing the preferred sharing method
Finally, Tester would have required reporting on how many indicators came in through DHS (clause I), how many came in through civilian agencies like FBI (clause II), and how many came in through military agencies, aka NSA (clause III). That would have provided a measure of how much data was getting shared in ways that might bypass what few privacy and oversight mechanisms this bill has.
Burr and DiFi replaced that with a measure solely of how many indicators get shared through DHS, which effectively sanctions alternative sharing.
That Burr and DiFi watered down Tester’s measures so much makes two things clear. First, they don’t want to count some of the things that will be most important to count to see whether corporations and agencies are abusing this bill. They don’t want to count measures that will reveal if this bill does harm.
Most importantly, though, they want to keep this information from Congress. This information would almost certainly not show up to us in unclassified form, it would just be shared with some members of Congress (and on the House side, just be shared with the Intelligence Committee unless someone asks nicely for it).
But Richard Burr and Dianne Feinstein want to ensure that Congress doesn’t get that information. Which would suggest they know the information would reveal things Congress might not approve of.
(E) A review of the type of cyber threat indicators shared with the Federal Government under this Act, including the following:
(i) The degree to which such information may impact the privacy and civil liberties of specific persons.
(ii) A quantitative and qualitative assessment of the impact of the sharing of such cyber threat indicators with the Federal Government on privacy and civil liberties of specific persons.
(iii) The adequacy of any steps taken by the Federal Government to reduce such impact.
Jon Tester amendment:
(I) the total number of cyber threat indicators shared through the capability described in section 5(c);
Section 5(c) is the DHS-routed intake on cyber information, with the scrub; it became 105(c) in the Manager’s Amendment. So this asks for how many threat indicators are coming through the officially-preferred means.
(II) a good faith estimate of the number of cyber threat indicators shared by entities with civilian Federal entities through capabilities other than those described in section 5(c);
CISA also permits companies to share with entities other than DHS, in which case it won’t get shared and scrubbed. I suspect this is a way of tracking how many threat indicators go straight to FBI, but there are other agencies that might get threat indicators directly.
(III) a good faith estimate of the number of cyber threat indicators shared by entities with military Federal entities through capabilities other than those described in section 5(c);
This is a polite way of asking how many threat indictors get shared with NSA directly.
(IV) the number of times personal information or information that identifies a specific person was removed from a cyber threat indicator under section 5(c);
This is a measure of how many specific persons’ data is getting scrubbed through the DHS process. It would measure the adequacy of any scrub on the private sector side, and the adequacy of the scrub at DHS (but would also require real numbers, which would require some effort).
(V) an assessment of the extent to which personal information or information that identifies a specific person was shared under this Act though such information was not necessary to describe or mitigate a cybersecurity threat or security vulnerability.
This is a measure of how many specific persons get sucked into this. The Intelligence Community surely hates it because it would provide a real number for how many Americans’ privacy got compromised with this, but also because it would require real auditing to determine. While CISA requires this process be auditable, nothing in it I see requires actual audits.
(VI) a report on any known harms by any defensive measure operated or shared under the authority of this Act’;
CISA permits private entities to fight back. This requires the IC to report on whether anyone has done damage with it. Again, I’m sure the IC hates it not just because it would require reporting on something they want to hide, but it would require auditing.
(VII) the total number of times that information shared under this Act was used to prevent, investigate, disrupt, or prosecute any offense under title 18, United States Code, including an offense under section 1028, 1028A, or 1029, or chapter 37 or 90 of such title 18; and
In addition to hacking, terrorism, and kiddie porn, CISA permits the use of CISA-derived data to be used in prosecuting identity fraud, espionage, and trade secrets. This requires the government to quantify how often it is not only used to prosecute such crimes, but also to investigate it, which is a far higher (but more appropriate) measure than FISA’s notice provisions to defendants.
(VIII) the total number of times that information shared under this Act was used to prevent, investigate, disrupt, or prosecute a terrorism offense under chapter 113B of title 18, United States Code.
As it says, this asks for the number of terrorists investigated or prosecuted using CISA data. The language permitting the use of CISA data for terrorism prosecutions is particularly squishy, permitting its use to “identify a cybersecurity threat involving the use of an information system by a … terrorist,” which I suspect may extend into terrorists on social media.
(E) A review of the type of cyber threat indicators shared with the appropriate Federal entities under this title, including the following:
(i) The number of cyber threat indicators received through the capability and process developed under section 105(c).
Note, they’ve specifically taken out the reporting on alternate reporting routes (and therefore routes that don’t include DHS’s scrub). All of those won’t get counted under this bill.
(ii) The number of times that information shared under this title was used by a Federal entity to prosecute an offense consistent with section 105(d)(5)(A).
This requires one number, all the prosecutions using CISA data. It wouldn’t tell Congress how much of this was actual hacking and how much other crimes. And it wouldn’t tell Congress how many people were investigated, but not charged, which in turn would make it less likely that DOJ would keep track of derivative prosecutions. That is, this would make it more likely DOJ would just parallel construct its way out of counting this.
(iii) The degree to which such information may affect the privacy and civil liberties of specific persons.
(iv) A quantitative and qualitative assessment of the effect of the sharing of such cyber threat indicators with the Federal Government on privacy and civil liberties of specific persons, including the number of notices that were issued with respect to a failure to remove personal information or information that identified a specific person not directly related to a cybersecurity threat in accordance with the procedures required by section 105(b)(3)(D).
Ron Wyden got an amendment added requiring the government to notify people if their information got inappropriately shared under this bill (though Burr and DiFi limited that notice to US persons; see F on page 13 for what it looks like). This requirement would apply to include 105(b)(3)(D), which mandates procedures for notifying federal agencies and private companies if something disseminated as a cyber threat indicator turned out not to be one. But all this refers back to the definition of cyber threat indicator, which is unbelievably broad, so notice to the feds and private companies, and therefore to the person herself, would only happen for stuff that didn’t fit that very broad definition.
(v) The adequacy of any steps taken by the Federal Government to reduce such effect.