A university got itself banned from the Linux kernel (2021)
theverge.com
theverge.com
[1] https://grants.nih.gov/policy-and-compliance/policy-topics/h...
This seems like one of those situations that would usually require regular review to err on the side of caution if nothing else. It's worth pointing out there are exceptions though:
https://grants.nih.gov/sites/default/files/exempt-human-subj...
Generally those exceptions fall into "publicly observable behavior", which I guess I could see this falling into?
It's ethically unjustified how the whole thing actually happened but I guess I can see an IRB coming to an exemption decision. I would probably disagree with that decision but I could see how it would happen.
In some weird legalistic sense I can also see an IRB exempting it because the study already happened and they couldn't do anything about it. It's such a weird thing to do and IRBs do weird things sometimes.
I mean I feel like the IRB is mostly dealing with medical stuff. "I want to electrocute these students every week to see if it cures asthma". "No that's too much.. every other week at most". "Great I'll charge up the electrodes"
So if a security researcher rolls in after the fact and says "umm yea so this has to do with nerd stuff, computers and kernels, no humans, and I just want it all to be super secure and nobody gets hacked, sound good" "ok sure we don't care if no people are involved and don't really understand that nerd stuff, but hackers bad and you're fighting hackers"
This story isn't evidence that IRBs are always useless, but also it's not an example of them being useful. The thing this story shows is they are sometimes useless.
Yes, they were. What kind of argument is this? If you submit a PR to the kernel you are explicitly engaging with the maintainer(s) of that part of the kernel. That's usually not more than half a dozen people. Seems pretty specific to me.
I assure you that it falls under IRB's purview -- I came into the thread intending to make grandparent's comment. When using deception in a human subjects experiment, there is an additional level of rigor -- you usually need to debrief the participant about said deception, not wait for them to read about it in the press.
(And if a human is reviewing these patches, then yes, it is human subjects research.)
Yes, if in the course of that experimentation, you also shipped potentially harmful products to buyers of those products "to see if Amazon actually let me".
There are cases where deception (as they call it) can be approved (even by ethics boards). Based on the Verge's article, this research setup should not have been approved even by then. But the topic itself seems as relevant as ever with the xz case and all.
I reported my advisor to university admin for gross safety violations, attempting to collect data on human subjects without any IRB oversight at all, falsifying data, and falsifying financial records. He brought his undergrad class into the lab one day and said we should collect data on them, (low hanging fruit!) with machinery that had just started working a few days prior, we hadn't even begun developing basic safety features for it, we hadn't even discussed design of experiments or requesting IRB approval for experiments. We (grad students) cornered the professor as a group and told him that was wildly unacceptable, and he tried it multiple more times before we reported him to university admin. Admin ignored it completely. In the next year, we also reported him for falsifying data in journal papers and falsifying financial records related to research grants. And, oh yeah, assigning Chinese nationals to work on DoD-funded work that explicitly required US citizens and lying to the DoD about it. University completely ignored that too. And then he got tenure. I was in a Top-10-US grad program. So in my experience, as long as the endowment is growing, university admin doesn't care about much else.
It's like if my wife said "I'm taking the car to get it washed" and then she actually takes the car to the junkyard and sells it. "Ha, you got fooled!". I mean, yes, obviously. She's on the inside of my trust boundary and I don't want to live a life where I'm actually operating in a way immune to this 'exploit'.
I get that others object to the human experimentation part of things and so on, but for me that could be justified with a sufficiently high bar of utility. The problem is that this research is useless.
1. Prof and students make fake identities
2. They submit these secret vulns to Greg KH and friends
3. Some of these patches are accepted
4. They intervene at this point and reveal that the patches are malicious
5. The patches are then not merged
6. This news comes out and Greg KH applies big negative trust score to umn.edu
7. Some other student submits a buggy patch to Greg KH
8. Greg KH assumes that it is more research like this
9. Student calls it slander
10. Greg KH institutes policy for his tree that all umn.edu patches should be auto-rejected and begins reverts for all patches submitted in the past by such emails
To be honest, I can't imagine any other such outcome could have occurred. No one likes being cheated out of work that they did, especially when a lot of it is volunteer work. But I was wrong to say the research was useless. It does demonstrate that identities without provenance can get malicious code into the kernel.
Perhaps what we really need is a Social Credit Score for OSS ;)
Research can be non-useless but also unethical at the same time...
> 4. They intervene at this point and reveal that the patches are malicious
> 5. The patches are then not merged
It's not clear to me that they revealed anything, just that they did fix the problems:
> In their paper, Lu and Wu claimed that none of their bugs had actually made it to the Linux kernel — in all of their test cases, they’d eventually pulled their bad patches and provided real ones. Kroah-Hartman, of the Linux Foundation, contests this — he told The Verge that one patch from the study did make it into repositories, though he notes it didn’t end up causing any harm.
(I'm only working from this article, though, so feel free to correct me)
You know what would really be wasteful of volunteer hours? Instituting a policy whereby the community has to trawl through 20 years of commits from umn.edu addresses and manually review them for vulnerabilities even though you have no reasonable expectation that such commits are likely to contain malicious code and you're actually just butthurt. (they found nothing after weeks of doing this btw)
Other than the tiny bit where that's not true. An institution just demonstrated that they are willing to submit malicious code, and don't feel any need to tell you that they did so (even after the fact). It's perfectly reasonable to ask if they've done this before.
Also, banning umn.edu email addresses didn't even make sense since the hypocrite commits were all from gmail addresses.
The blanket ban was kicked off by another incident after the hypocrite commit incident.
2) I don't think it's unethical to send someone an email that has bad code in it. You shouldn't need an IRB to send emails.
It's unethical because of the bits you left out: sending code you know is bad, and doing so under false pretenses.
Whether or not you think this rises to the level of requiring IRB approval, surely you must be able to understand that wasting people's time like this is going to be viewed negatively by almost anyone. Some people might be willing to accept that doing this harm is worth it for the greater cause of the research, but that doesn't erase the harm done.
GKH's response was to waste man weeks or man months of maintainer time persecuting every last commit that happened to come from umn.edu, despite having zero reason to believe these commits were more suspect than any other institution's commits.
It's totally possible to obtain evidence of that without being an asshole to kernel maintainers. Which is the kind of thing that an ethics review conducted before the experiment could have pointed out. If the goal of the experiment was merely to demonstrate the lack of routine static analysis capable of catching such vulnerabilities, then the experiment's design was not justified and the experiment was needlessly harmful to non-consenting participants.
2) Yes, emails absolutely need IRB sign-off too. If you email a bunch of people asking for their health info or doing a survey, the IRB would smack you for unapproved human research without consent. Consent was obviously not given here.
[0] https://lore.kernel.org/linux-nfs/CADVatmNgU7t-Co84tSS6VW=3N...
2) I don't actually care about the internal policies of UMN's IRB. Whether or not the study's approval was proper and whether they would get into trouble with their boss is not my problem. The point is that what they did is obviously not immoral or unethical.
The "ethical" issues with this study do not rise to the level that I care, so the only objection is that they didn't get the IRB to rubber stamp it beforehand, which I also don't care about.
I believe most people believe that the Linux kernel couldn’t be compromised because there is multiple approval process and highly professional people vetoing.
It seems like a big vulnerability, if a teacher assistant could do that, there is no doubt that government agencies can too.
I had to apply for exemptions often in grad school. You must do so before performing the research -- it is not ethical to wait for outcry then apply after the fact. Any well run CS department trains it's incoming students on IRB procedures during orientation, and Minnesota risks all federal funding if they continue to allow researchers to operate in this manner.
(Also "exempt" usually refers to exempt from the more rigorous level of review used for medical experiments -- you still need to articulate why your experiment is exempt to avoid people just doing whatever they want then asking for forgiveness after the fact)
This level of malfeasance strikes me as something akin to plagiarism for a professional writer.
More to the point; are they salty because the author has possibly proved that it's most certainly possible to get critical flaws into the Linux kernel with social engineering? How else is something like that meant to be tested?
If you give them a heads-up they'll pay more attention for a short duration of time.
But there is always the BSDs.
I found doing that makes maintenance far easier that what I had to do with html.
Money is money and buys time, no harm done, useful research conducted, and a whole lot of publicity gained.
That says a lot about Linux kernel safety.