A few months back I fact-checked the CSIRO CEO’s evidence to a Senate Estimates Committee hearing. At the time I promised more fact-checking of CSIRO’s communication about the abandoning of Trustworthy Systems (TS). Well, it has taken much longer than I hoped, mostly because I’ve been working 60-hour weeks to undo the damage caused by CSIRO to the TS group, and seL4 and its ecosystem. Six months down the track I’m winning, but it’s taking all my time and energy.
So, somewhat delayed, here we go…
CSIRO Statements
There are at least six ways in which CSIRO communicated about the dismantling of TS:
- Statements made by management to all Data61 staff
- Statements made by management/HR to TS staff
- Statements made by management to unions
- Statements made by CSIRO to the media
- Statements made by CSIRO to industry partners
- Statements made by the CSIRO CEO to the Australian Parliament
Let’s look at them in turn.
1. What CSIRO told all Data61 staff
Exactly 6 months ago, on 17 May 2021, CSIRO’s Data61 held an all-staff meeting where management informed staff about the new Data61 strategy. While there were some extremely vague statements about the new strategy earlier in the year, this was the first time things became concrete.
Some of the statements made there by management are (and I point out that none of the following are literal quotes, as I do not have access to a recording/transcript or the slides presented, only my sparse notes):
- Data61 will be doing “fewer, bigger things”
- Data61 will be all about AI
- Data61 will continue to do cybersecurity, but only
- cybersecurity for AI, and
- AI for cybersecurity
- Data61 will exit “less aligned areas”
- There will be “minimal impact” on university collaborations (so-called CRP agreements)
We’ll refer to those statements as S1.1–S1.5 in the following.
2. What CSIRO told TS staff
On 20 May, TS staff received email from CSIRO HR stating (among others):
In relation to the Trustworthy Systems Group, the review has identified that while this group has achieved impact in the past, the capability area is not aligned with the three new strategic goals of the Data61 Strategy and fewer and bigger things identified. In particular, we will be focusing on the intersection of AI and Cyber, not just any cybersecurity and related research.
This literal quote barely qualifies as English, but let’s try to separate it into individual statements (referenced as S2.1–S2.3 below):
- the capability area is not aligned with the three new strategic goals of the Data61 Strategy
- [TS is not aligned with doing] fewer, bigger things
- [Data61] will be focusing on the intersection of AI and Cyber, not just any cybersecurity and related research
3. What CSIRO told unions
On 21 May, CSIRO management sent an Advice to Unions, stating (among others):
Data61 will pivot away from areas that lack critical mass, are not at the cutting edge scientifically, and/or do not have sufficient market demand.
Separating the sentence into individual claims S3.1–S3.3, we get:
- pivot away from areas that lack critical mass
- [pivot away from areas that are] not cutting edge scientifically
- [pivot away from areas that] do not have sufficient market demand
On 21 May, a CSIRO spokesperson was quoted as:
The Trustworthy Systems group is focused on the area of formal methods for design, implementation, and verification of software systems. It is [sic] mature area of technology that CSIRO has invested in over a number of years and is now well supported outside the organisation
https://www.innovationaus.com/data61-dumps-world-class-sel4-security-team/
We separate the claims into:
- [what TS is doing] is a mature area of technology
- [TS technology] is now well supported outside the organisation
5. What CSIRO told industry partners
An email sent to industry partners (which reached me at my UNSW address via third parties not encumbered by NDAs) said:
- Data61 is committed to deliver to our partners, such as [company], especially on current projects and any work-in-progress.
6. What CSIRO’s CEO told parliament
In my earlier blog I fact-checked the testimony Dr Marshall, CSIRO’s CEO, gave to the Australian Senate. The conclusion was that some of it was factually incorrect and relevant information was omitted (and my clarifications/corrections to Dr Mashall were not passed on the parliament). I won’t go through all of it again, but am extracting the bits that are relevant for contrasting other statements made by CSIRO:
- [TS technology is] very mature and it’s open source, so it’s difficult to see an opportunity to build an industry in Australia or to derive a national benefit from that technology.
Let’s examine these statements in turn
Statements to all staff
S1.1: “Data61 will do fewer, bigger things.”
TS was unique inside Data61 in tackling large problems with critical mass and a unique combination of skills. Our strategy explicitly said for years “we solve problems no-one else can” – and we did.
Assessment: This is exactly what TS is known for, so that’s not a reason for destroying TS, to the contrary.
S1.2 + S1.3: Data61 will be all about AI, plus cybersecurity as long as it’s for AI systems.
It’s obviously management’s prerogative to define what to include or exclude in their research strategy (if you can call a bunch of buzzwords a “strategy”). However, “cybersecurity for AI systems” would clearly include the work of TS. Here are a few proof points:
- The DARPA HACMS and CASE programs explicitly use TS technology to protect autonomous systems from cyber-attacks.
- seL4 Foundation member Ghost stated: “There is more research required into how to architect, construct and integrate an AI application on a trusted system of this complexity, and how we achieve this at scale. Investing in AI research without investing in trustworthy systems research will greatly diminish the impact and applicability of AI to real-world products.” Clearly, TS technology ensures cybersecurity of AI systems.
- Another seL4 Foundation member, Horizon Robotics, said: “To address the challenges of safety, security and realtime in autonomous software, a fundamental high quality state-of-the-art microkernel is needed. We are looking forward to working with members of seL4 Foundation to build mixed-critical platform and solution for next-generation autonomous driving vehicles.” Again, TS technology providing cybersecurity for AI systems.
There are more examples of this. Clearly, if you’re serious about cybersecurity for AI systems, then TS’s is the go-to technology, so ruling TS out of scope on that base is dishonest or clueless.
But it gets worse. Data61 seems to keep investing in blockchain technology. There’s nothing inherently wrong with that, but how does it fit the criteria that supposedly rule TS research out of scope? Blockchain may be a popular buzzword, but it’s not AI. And even calling it cybersecurity for AI is a huge stretch. It’s much less obviously aligned with the Data61 “strategy” than TS technology is. The same can be said about Data6’s IoT-Cloud Security work.
Assessment: Ruling TS technology out – while ruling blockchain and IoT-Cloud security research in – based of the Data61 strategy is inconsistent and dishonest.
S1.4: Data61 will exit “less aligned areas”
As above: TS research is clearly aligned with “cybersecurity for AI systems”, and is better aligned than blockchain research and at least as well as “IoT-Cloud Security”, yet they “exit” TS but, apparently, not the others.
Assessment: inconsistent and dishonest.
S1.5: There will be “minimal impact” on university collaborations (CRPs)
Well, the TS CRP with UNSW was one of the biggest (if not the biggest). It was terminated early as a consequence of the “exiting” of TS. Do you call this “minimal impact”?
Assessment: Looks pretty dishonest to me.
Statements to TS staff
S2.1: [TS] is not aligned with the three new strategic goals of the Data61 Strategy
As explained above, TS is actually aligned with “cybersecurity for AI”, as independently evidenced by external sources. Yet, the clearly less-aligned blockchain area seems to continue.
Assessment: dishonest.
S2.2: [TS] is not aligned with “fewer, bigger things”
Assessment: debunked above (S1.1).
S2.3: we will be focusing on the intersection of AI and Cyber, not just any cybersecurity and related research
Assessment: debunked above (S1.3).
Statement to Unions
These were a number of “and/or” conditions, so notionally the combination was true if at least one of them was true. Let’s look at them one by one.
S3.1: exiting areas lacking critical mass
TS used to be the by far biggest group in Data61, there’s no doubt there was critical mass. By the time of the decision, this had been withered down to about 14 CSIRO full-time staff by not replacing departures and not renewing fixed-term contracts (leading to projects being understaffed and failing to deliver on time – partners getting screwed as a result). In addition there were about 6 FTE UNSW research and engineering staff. 20 FTE is still a pretty good group size, so this claim does not hold for TS.
S3.2: exiting areas “not cutting edge scientifically”
There is plenty of work on-going in Data61 that is nowhere near “cutting edge scientifically”, but TS wasn’t one of them. Just to give one example, our work on time protection as a principled means for completely eliminating information leakage through timing channels is without doubt cutting-edge, I’d even claim it’s peerless. Just one indication is that out of 4 publications in the 2 years prior to the decision to abandon TS, that work has won three best-paper awards. I challenge anyone to publicly state that this work is not cutting edge – they’ll open themselves up to ridicule. So, this claim also doesn’t hold for TS.
S3.3: exiting areas that do “not have sufficient market demand”
In the context of TS, this one is directly contradicted by CSIRO’s own public statements, specifically S4.2, as well as the strong uptake of seL4 and the growth of the seL4 Foundation in the past 6 months. So, this claim doesn’t hold for TS.
Assessment: neither of these individual claims hold for TS, and hence using them (even in combination) as a justification for “pivoting away” from TS research is dishonest.
The astute reader will notice that the complete quote avoids associating TS with cybersecurity in any way. That in itself is interesting, given that the importance of cybersecurity is much more widely understood than that of software verification, even though you can’t have real cybersecurity without verification. This looks very much like spin.
But let’s look at the individual statements:
S4.1: TS technology is mature and doesn’t require on-going support
I wrote a whole blog debunking this nonsense. I’m not going to repeat it here, but the summary is while seL4 is mature enough to be deployed in the real world, there’s plenty of fundamental research work left on seL4 itself (such as the award-winning time protection work), and there is far more research left on how to achieve real-world trustworthy computer systems.
Assessment: Utterly wrong.
Statements to industry partners
S5.1: Data61 is committed to deliver to our partners
So, they claimed they are “committed” to delivering to customers, just as they decided to get rid of all the people who have the skills for such delivery? Making such a statement to partners who are handing over large sums to CSIRO in exchange for getting research done which CSIRO is now obviously incapable of delivering is just stunning. And management can’t pretend they didn’t know that, the uniqueness of TS’s skillsets are widely known in the organisation. It’s a shame none of the affected partners sued them, they would have had a good case I reckon.
What happened in the end is that one partner terminated their multi-million-dollar project (after having handed over the majority of that cash). The others we transferred to UNSW, where they are served by the remaining TS team. However, that possibility only existed because UNSW offered to fund the team to the end of the year – without that offer (which was made after CSIRO’s announcement, of which UNSW was given no prior warning) there would be no TS team left. So if CSIRO were to make any claims (which were at least implied in some of the statements) that all is good because we’re continuing at UNSW that would add to the dishonesty.
Assessment: Blatantly dishonest.
Statements to the Australian parliament
I already fact-checked Dr Marshall’s evidence to the Estimates Committee of the Australian Senate, and concluded that it omits important facts, and especially with regard to a senator’s question about work on protecting autonomous cars from cyber-attacks (of which CSIRO has none after demolishing TS). I pointed out that the claim of “difficult to see an opportunity to build an industry in Australia or to derive a national benefit from that technology” cannot be reconciled with the fact that there is a growing number of Australian companies whose business is based on TS technology. (And to my knowledge, there is no research remaining on Data61 that can make a similar claim.)
A further dis-proof of this claim is the Laot project, funded by Australian Defence, which deploys TS technology for protecting Australia’s critical infrastructure. And silly me thinks that would be in the national interest? Particularly after Australian Home Affairs Secretary Mike Pezullo’s warning about attacks on critical infrastructure?
Moreover, Dr Marshall failed to correct the record after I provided the information to him, so he has no place to hide.
Assessment: Omits important facts.
Comparing CSIRO statements
Amazingly, CSIRO’s statements aren’t even consistent with each other. Specifically, CSIRO makes claims that TS technology:
- “does not have sufficient market demand” (S3.3);
- “is now well supported outside the organisation” (S4.3);
- “[no] opportunity to build an industry in Australia or to derive a national benefit” (S6.1).
I can’t see how to reconcile these statements with each other, especially “no market demand” on one side and at the same time “well-supported outside” (by industry!)
Assessment: contradictory.
Conclusions
The seemingly inescapable conclusion from the above is that CSIRO’s communications about TS are full of half-truths, untruths and contradictions. This is a pretty sorry state of affairs for our national science agency, and specifically its Data61 arm that is all about the future-critical information technology. Note that this is an organisation that lists “trusted” as one of its values. How can you trust an organisation that is so loose with the truth?
Unfortunately, this is consistent with the long and slow decline of CSIRO, resulting from budget cuts and political interference. I am actually amazed that there is still great science being done in parts of CSIRO, despite the oppressive culture (likened by some to the Soviet Union) created by management. Data61, despite all the propaganda about excellence etc, based on my observations (and that of others), is an organisation that fosters mediocrity and is hostile to excellence.
Four years ago, Data61’s then Chief Scientist Prof Bob Williamson FAA conducted an internal review of Data61 science. It identified 6 groups/teams as being top-class, TS being one of them. (Another one was the Legal Informatics team that got disbanded at the same time as TS.) Not much is left of those, the top people left or their teams got abolished. What is left is mostly mediocre.
The same review also identified areas that were sub-par. It was particularly scathing about one team:
This is a weak team that does not know it. Unfortunately they appear to have been favoured by the program director who has a very poor sense of scientific quality. The interview and documentation was disappointing. They are actually a reputational risk to the organisation. They should be disbanded. I doubt there is talent here that is worth retaining.
That team is continuing in Data61, and its leader has been given extra responsibilities. It’s all you need to know.
Data61 management is likely to counter this with the external “science review”, which contained a fair amount of uncomplimentary material, some of which was apparently aimed at TS, especially comments about living off the past. The problem is that we were supposed to talk only about impact, of which TS has plenty, but, of course, that’s inherently a rear-mirror view. Any reference to our on-going research and vision was censored from the group leader’s talk by management and comms people! Talk about setting world leaders up for failure!
My final conclusion is that Data61 is an institutionalised mediocrity and a waste of taxpayers’ money. Eventually, people will realise this and shut it down (or completely revamp it). While inevitable, this may take up to five years – plenty more tax dollars will be wasted until then. Depressing prospects for the tax payer.
But on the bright side: TS is over the hill and is regrowing its strength. We aren’t done yet with producing excellence and impact – stay tuned!
from Hacker News https://ift.tt/3DCBzQc
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.