When FERC mandated in July 2016 that NERC develop a supply chain
cybersecurity risk management standard, just about everyone (including me)
focused on the fact that it was probably the first supply chain cybersecurity
standard outside of the military – in the US and possibly anywhere. This was
true, but NERC CIP-013 (produced in response to FERC’s order) was also one of
the first cybersecurity risk management standards: i.e., a standard
whose explicit goal was establishment of a cyber risk management program. I
think that will be CIP-013’s most important legacy. This is why I say that:
A
brief, admittedly biased history of the CIP standards
No NERC standards before CIP
version 1 had even mentioned anything about risk. And why should they have?
They were all about physical actions: trimming trees under Transmission lines, balancing
load and supply in real time, etc. The results of these actions could be
objectively measured and needed to be kept within certain operating limits. Risk
management had nothing to do with those standards. They were based ultimately
on the laws of physics, which in themselves don’t understand risk.
However, the team that drafted CIP
version 1 knew that cybersecurity was different. They knew there’s just about
zero historical data that will let you make predictions like, “If we decrease
our patching interval from 45 to 30 days, our chances of suffering a cyberattack
– that could cause an adverse grid event – will go down by 15%.”
Instead, the only thing you can
say with certainty is, “Patching more frequently will in general lower the risk
that we will be compromised through an unpatched software vulnerability.” Does
that mean a utility should patch every hour, even if it means shutting down the
rest of its operations and putting its entire staff to work on patch
management? No. Like every organization, electric utilities have limited
resources and have many other risks (both cyber and otherwise) that they need
to manage. They need to balance their various risk management activities
against each other, so they can reduce as much overall risk as possible,
given their available resources. Oh, and they have to keep the lights on at the
same time.
In other words, cybersecurity is
at heart about risk management. An important principle of risk management is accepting
risks that are too expensive to mitigate, when weighed against the benefits
that would be realized by mitigating them. Sure, a meteor strike on your
headquarters would have a devastating effect on an organization (to say nothing
of their people). Does that mean that every organization in the world should
spend every extra dime (shekel, euro, rupee, etc.) they have on protecting
their headquarters against a meteor strike?
No, because risk is a combination
of likelihood and impact. The impact of a meteor strike would be huge, but the
likelihood is so small that the risk itself is close to zero. No organization that
I know of spends anything to fortify their headquarters (or anything else)
against meteor strikes. By the same token, there’s no amount of spending on
cybersecurity that would allow any organization – electric utility, dry
cleaners, pizza parlor, the US military, etc. – to say it was perfectly cybersecure.
There ain’t no such thing. There will always be some residual risk that the
organization needs to accept.
For this reason, the CIP v1
drafting team included in many of the CIP v1 requirements the words “…or
acceptance of risk…” In other words, if the utility believes that the cost of
fully complying with the words of the requirement would outweigh whatever
benefits (in risk reduction) would be achieved by compliance – and can document
their reasons - they would have the option of not fully complying with those
words. And they wouldn’t be held in violation of the requirement.
But FERC wouldn’t have any of
this. When they approved CIP v1 – after 17 months of consideration – in early
2008, they said, somewhere in their 800 (or so)-page Order 706, that they
wanted NERC to start work on a new version that would, among many other things,
eliminate the wording about acceptance of risk. And in the meantime, NERC
needed to audit as if those forbidden words weren’t there (by the way, if you
want to see a very impressionistic history of NERC CIP that I wrote in 2018, go
here).
But this caused a problem: The team
that drafted CIP v1, believing that the language about acceptance of risk would
remain in the requirements they were drafting, deliberately made the
requirements quite prescriptive. After all, if a NERC entity found they couldn’t
follow the prescriptive wording of the requirement, they could always just
accept the risk. What could possibly go wrong?
As it turns out, a lot. By eliminating “acceptance of risk” but not changing the
overly prescriptive nature of the CIP v1 requirements, FERC left NERC with the
worst of both worlds: hard-edged prescriptive requirements with no means of “softening”
them through risk considerations.
Over the first four or five years
of CIP enforcement – and as CIP v1 was replaced by v2 and then v3 - there were
loud and growing complaints from NERC entities about the amount of time and
money they were being required to spend to comply with rigidly prescriptive
requirements. Yea, great was the weeping and wailing and gnashing of teeth over
NERC’s “zero-tolerance” (i.e. non-risk-based) enforcement of the CIP
requirements.
It wasn’t until CIP version 5 came
into effect in 2017 that there were any CIP requirements (let alone entire
standards) that at least implicitly allowed NERC entities to take risk into
account: these pioneering requirements, both part of CIP v5 (which was a
complete rewrite of all the CIP standards), were CIP-011-5 R1 (Information
Protection) and CIP-007-5 R3 (Anti-Malware)[i].
But the watershed event in moving
to a risk management approach was FERC Order 829, which ordered NERC to develop
a supply chain security standard in July 2016. Even though NERC had never –
since their unfortunate experience with “acceptance of risk” - officially
referred to “risk” in any standard, in Order 829, FERC specifically called for
a “supply chain (cyber) risk management” (my emphasis) standard. And
they specifically warned against “one size fits all” – i.e. prescriptive –
requirements.
Why this change of heart? For one
thing, FERC certainly had heard the complaints about the then-current standards
(when they issued Order 829 in 2016, the industry was still complying with CIP
version 3, although they were preparing for CIP v5, which came into effect on
July 1, 2017), and knew that the last thing NERC entities needed was another
highly prescriptive CIP standard.
But I believe (and believed in
2016) that the main reason why FERC wanted the new standard to be risk-based
was because there was simply no other way to do it. Remember, even though the
burden of compliance with CIP-013 falls on electric utilities (and other bulk
electric system entities like federal power marketing agencies and independent
power producers), the standard is really aimed at the suppliers of the
hardware, software and services that those utilities rely on to operate the
BES.
Let’s say a utility decided to
hold all of those suppliers to a very high standard of cybersecurity, e.g. ISO
27001. For some suppliers, such as the supplier of the energy management system
(EMS) that literally runs the utility’s own corner of the grid, it makes sense
to require this. However, for other suppliers, such as perhaps the vendor of
maintenance services in a power plant, this would be overkill.
If the utility tried to force the
maintenance services vendor to comply with ISO 27001, they would quickly find
themselves looking for a new vendor. A vendor isn’t going to spend many times
the profit they may make from a customer in order not to lose that customer;
they’ll simply try to find another customer (perhaps in another industry) that
isn’t so demanding. And if the vendor can’t find another customer to replace
the utility, they would still be much better off financially than if they’d
agreed to pay for an ISO 27001 audit just to keep one customer.
And this is the problem with
supply chain risk management: The organization has to convince their vendors to
incur costs in order to keep them as a customer. They’ll never do that if they require
them all to comply with the same high standards, rather than tailor their
requirements to the degree of risk posed by each vendor. A supply chain risk
management standard like CIP-013 has to be risk-based, if it is to have any
chance of succeeding.
The NERC standards drafting team
(SDT) took Order 829 to heart when they developed CIP-013. In fact, in my
opinion, the SDT went a little too far. CIP-013 consists of a grand total of
five sentences (although with a number of clauses), divided into three requirements:
1.
The first part of requirement
R1 (R1.1) tells the utility to develop a “supply chain cyber security risk
management plan(s)”. The plan needs to “..identify and assess cyber security
risk(s) to the Bulk Electric System from vendor products or services resulting
from: (i) procuring and installing vendor equipment and software; and (ii)
transitions from one vendor(s) to another vendor(s)..”
2.
The second part of R1 (R1.2)
identifies six risks that need to be included in the plan, such as “Disclosure
by vendors of known vulnerabilities related to the products or services
provided to the Responsible Entity”. While all six of these risks are important
ones, they weren’t developed as some sort of comprehensive catalog. Rather, the
drafting team (and I attended a number of their meetings) simply gathered in
one place six statements about particular risks that FERC made at various
random points in Order 829. These six risks are by no means the only six that
need to be included in the utility’s supply chain cyber risk management plan.
But these are the only risks that FERC specifically required to be included in
the plan; the utility is free to decide for themselves what other risks should
be included.
3.
R2 requires the utility
to “..implement its supply chain cyber security risk management plan(s)
specified in Requirement R1.” That’s literally all it says.
4.
R3 requires the
utility to review its plan every 15 months.
I just summarized the entirety of
CIP-013. The utility has lots of freedom to develop their supply chain cyber
risk management in R1, but in R2 they have no freedom at all: they have to
implement the plan as written. However, this isn’t as bad as it sounds: If the
utility decides they made a mistake in their original plan – or if they realize
that changed circumstances require a change in the plan - they’re free to
change it at any time; they just have to document the changes they made and why
they made them.
When NERC entities were starting to
think about CIP-013 compliance, I know some of them made a very understandable
mistake (especially for NERC entities): They focused on the six items in R1.2
and considered these to be the total of what’s required by CIP-013. After all,
those six requirement parts were the most like the existing CIP requirements; why
shouldn’t they be the only real “requirements” in CIP-013?
However, taking that attitude was based
on completely ignoring the wording at the beginning of CIP-013 R1: “Each
Responsible Entity shall develop one or more documented supply chain cyber
security risk management plan(s)… The plan(s) shall include:..” This is
followed by the text of R1.1 and R1.2, meaning that what’s described in both R1.1
and R1.2 needs to be included in the plan. So, without a doubt, the plan needs
to include the six R1.2 items. But it also needs to include R1.1.
What threw these utilities off was
probably the fact that R1.1 doesn’t prescribe anything in particular: It just
requires the entity to develop a plan to “identify and assess” supply chain
cybersecurity risks. In other words, it’s up to the entity themselves to determine
what are the risks they face. Of course, that idea was a 180-degree change from
all previous NERC standards. If you were complying with any of the CIP version
3 standards, or the FAC, BAL, etc. standards, and you told an auditor that you should
be allowed to decide what goes in your compliance plan, not NERC, how far do
you think that would get you?...You’re right – not very far at all.
For this reason, it’s
understandable that NERC entities wouldn’t know what to do with a requirement
that says it’s up to them to decide what goes into their supply chain cyber
risk management plan. What’s to prevent an entity from simply including the six
R1.2 items in their plan? In other words, how could they be found non-compliant
if that’s all they put in their plan?
They can be found non-compliant
because they’d have to convince the auditor, as part of their R1.1 compliance
evidence, that they searched high and low to find any other supply chain cyber
risks that applied to them, beyond the six risks in R1.2 – and they just
couldn’t find any others at all. If I were the auditor and that evidence were
prevented to me, I’d just ask some questions like:
1.
What about
SolarWinds-type attacks? Don’t you think you should be concerned about those?
More importantly, what have you done that makes you immune to such attacks?
2.
What about Log4j-based
attacks? Have you determined that you don’t have any log4j at all in your
environment (even in a component of a component of a component)?
3.
How about the risks
identified in the NATF Criteria? They include the six R1.2 items, but they go
well beyond those (there are 60 Criteria now), including:
a.
The risk that a
software or device supplier won’t follow a secure development lifecycle (criterion
#47);
b.
The risk that a
supplier might install a backdoor while developing a product and not remove it
before they ship it to you, leading to your being compromised (criterion #15);
c.
The risk that a
product supplier might not conduct 7-year background checks on its employees (criterion
#4); etc.
These are all supply chain cyber
risks. The entity will have to convince the auditor that they at least
considered all of these risks when they were developing their plan. And if they
didn’t include them in the plan, they’ll have to provide some justification for
not including at least each of the 60 NATF Criteria. From what I’ve heard about
how CIP-013 is being audited, and from the presentations I’ve seen by regional
auditors on the subject, I really don’t think an entity that only included the
R1.2 risks in their plan wouldn’t find the auditor to be fairly
skeptical.
In retrospect, it would have been
good if the drafting team had included a list of say ten areas of supply chain
security risk that the entity needs to consider in their plan; if they decide one
or more of those areas don’t apply to them, they need to document that fact.
These areas might include:
1.
Software supply chain
risks, including the risk that a malicious party might have implanted a
backdoor in the software while it was being developed, as in SolarWinds; or the
risk that a serious vulnerability would be identified in a widely used open
source component like Log4j, making it difficult even to find all the
vulnerable instances on your network.
2.
Inadequate protection
of the supplier’s remote access system (i.e. no MFA). DHS said
in 2018 that at least 200 vendors to the power industry had been penetrated by
the Russians through their remote access systems, in attempts to penetrate US
electric utilities through them.
3.
Inadequate
anti-phishing training and other anti-phishing measures, making the vendor a
possible vector for attacks aimed at the utility. In early 2019, the Wall
Street Journal published a great article
on how the Russians were penetrating vendors to the power industry through
phishing attacks, then utilizing that access to penetrate electric utilities;
the article listed four utilities that had been penetrated this way.
Had the drafting team included
this list in R1 (and I certainly never even thought to suggest it to them), this
would have made it clear to utilities that the purpose of R1 was to allow them
to decide the risks that were most important to them, given their configuration
of assets and vendors. Being able to allocate your resources toward the risks
that are most important is a key element of supply chain cyber risk management.
And there would have been another
benefit: Including this list in the requirement itself would have made it
auditable. That is, the auditors would have been able to determine whether or
not the utility gave serious consideration to each of these areas of risk,
based on the documentation they were shown. If the utility hadn’t even considered
each of these areas, they would have been eligible for a PNC (potential non-compliance)
finding in an audit.
However, as far as I know, a
substantial majority of NERC entities (who have medium or high impact BES Cyber
Systems, since they are the only ones in scope for CIP-013 so far) did a good
job of identifying and assessing supply chain cyber risks in R1, in spite of there
not being a list of risks in the requirement itself. This is because
organizations – especially the North American Transmission Forum (NATF) –
stepped up to provide the guidance that the drafting team didn’t want to
include in the requirement itself.
To get back to the question in the
title, has CIP-013 worked? Yes, it has. It’s worked both as a supply chain
cybersecurity standard, and as a cybersecurity risk management standard. It
could have been better, but it could also have been a lot worse.
Any opinions expressed in this blog post are strictly mine and are not necessarily shared by any of the clients of Tom Alrich LLC. If you would like to comment on what you have read here, I would love to hear from you. Please email me at tom@tomalrich.com.
[i] They
were joined by CIP-010-2 (now -3) and CIP-003-7 (now -8), which are also risk-based
requirements – although neither of them mentions risk, either.
No comments:
Post a Comment