bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 3쪽 중 1쪽
p. 01
Version 15
A Decentralized
Cyber Threat
Intelligence Market.
p. 02
Version 15
Table of Contents
PolySwarm in 60 Seconds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 03
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 04
Reinventing the Threat Intelligence Market . . . . . . . . . . . . . . . . . . . . . . . . . . 06
The Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 07
Prediction Markets, Arbiters & Mediated Consensus . . . . . . . . . . . . . . . . . . 09
Determining Ground Truth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
The Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Rewarding Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Fees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Protocol Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Bounty Lifecycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Offer Lifecycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Reputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Worker Registry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Artifact Confidentiality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Additional Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Token Sale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Disclaimer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
p. 03
Version 15
PolySwarm in 60 Seconds
Polyswarm is a decentralized threat intelligence market
made possible by Ethereum smart contracts and blockchain
technology .
Polyswarm incentivizes rapid innovation in the $8 .5B/yr anti-virus and
automated cyber threat intelligence space with precise economic incentives that
reward timely and accurate threat intelligence concerning the malintent of files,
network traffic and URLs .
PolySwarm defines a real-time threat detection ecosystem involving
enterprises, consumers, vendors and geographically-diverse security
experts. Experts develop and hone competing “micro-engines” that
autonomously investigate the latest threats, attempting to outperform
their competition. PolySwarm’s “Proof of Work” is threat detection
accuracy: the market rewards experts who are best able to defend
enterprises and end users.
Relative to today’s ad hoc market, PolySwarm will lower the barrier
to entry, provide broader coverage options, discourage duplicative
effort and ensure interoperability among products and threat
intelligence feeds.
Economically, PolySwarm functions as a skill-required twist on a
prediction market2 with thousands of micro-engines (“workers”)
investigating the latest in malware evolution at machine speed - no
human in the loop.
PolySwarm will be developed by PolySwarm Pte. Ltd. with funding
derived from the sale of ERC20-compatible Nectar (“NCT”) utility
tokens.
As a utility token, PolySwarm economically disincentivizes Nectar
speculation by rewarding honest market participation through the
collection and distribution of Fees (details on page 6) to value-
adding, active participants.
1 Jefferies Cyber Security Primer. January 18, 2017.
2 https://en.wikipedia.org/wiki/Prediction_market
p. 04
Version 15
Background
Today’s enterprises rely on an ad-hoc mixture of anti-virus subscriptions,
threat intelligence feeds, and assorted dynamic analysis engines to defend
against evolving adversarial cyber activity . Users must weigh the benefits and
drawbacks presented by each solution and decide on the least-worst fit for
their environment .
Today’s market discourages solutions that provide broad threat
coverage.
Today’s solutions focus on a comfort zone of threats - a direct
result of today’s market economics.
Figure A: The black rectangle represents all the threats an Enterprise may encounter; white
circles are anti-virus product #1 and #2, respectively.
It’s easy (and perhaps justifiable) to ignore an anti-virus solution
that doesn’t detect WannaCry3, but in doing so, today’s market
effectively rewards overlapping coverage among vendors - a market
inefficiency that causes duplicative cost. This is a classic tragedy of
the commons situation.
Similarly, consider a vendor that chooses to develop expertise outside
of this comfort zone. If the comfort zone is defined by the set of
threats faced by most enterprises, sales for this specialized vendor
will be difficult. How do you convince a potential customer that they
3 https://en.wikipedia.org/wiki/WannaCry_ransomware_attack
p. 05
Version 15
will have to deal with malware that you are uniquely qualified to
detect / prevent / mitigate?
Finally, defenders cannot mix and match many of today’s solutions,
making combinatorial coverage impossible in many scenarios.
In contrast, PolySwarm will foster an ecosystem of broad coverage
powered by 1000s of “micro-engine” workers, authored by
geographically diverse security experts.
Figure B: PolySwarm will foster an ecosystem that produces broad threat coverage.
PolySwarm provides a legitimate revenue stream for security
experts.
When local economics cannot support sufficient honest work, some
security experts develop ransomware, operate bots and otherwise use
their skills for evil. PolySwarm provides a region-free alternative:
unencumbered reward for honest work. Experts compete to make the
internet a safer place.
PolySwarm puts the user first.
The PolySwarm market directs economic incentives to where it matters
most: toward accurate malintent detection (“threat intelligence”).
Accuracy is determined via a novel process we call Mediated
Consensus (details on page 10).
In short, PolySwarm provides users with timely access to broad,
crowdsourced security expertise. Need to quickly triage a suspect
file? Just Swarm ItTM.
p. 06
Version 15
Reinventing the Threat Intelligence
Market
PolySwarm's Nectar ("NCT") tokens form the basis of a new market that
introduces novel instruments for satisfying demand for timely and accurate
assertions regarding the malintent of files, network traffic, and URLs,
collectively referred to as Artifacts . These new instruments are structured
to directly incentivize innovation in the threat intelligence space with a
feedback loop driven by accurate results .
PolySwarm employs Fees to discourage spam and incentivize honest
and active market engagement. Fees are assessed on transaction
types that may be abused for spam, and then distributed to active
ecosystem participants; those who are introducing Artifacts and
determining ground truth regarding the malintent of Artifacts. This
participation is measured in a sliding window fashion, “ageing-off”
older market contributions, thereby incentivizing continued market
participation.
PolySwarm 1.0 (and this document) will focus exclusively on boolean
(malicious / benign) determination, but the PolySwarm team has
bigger plans than disrupting boolean malintent determination. During
PolySwarm 1.0 development, the team will investigate methods to
incentivize the production of artifact metadata such as malware
family and how core PolySwarm concepts such as smart contract
market design, Mediated Consensus and an information security focus
can disrupt related markets such as vulnerability bug bounties4.
Before introducing PolySwarm’s instruments, we introduce the
participant classes that will utilize them.
Enterprises Experts Ambassadors Arbiters
4 “HackerOne but distributed, pseudonymous, and unfettered by jurisdictional
encumbrances. Powered by blockchain.” Not covered by this document.
p. 07
Version 15
The Participants
END USERS
AMBASSADOR
(INFOSEC COMPANY)
NO REWARD
(ENTERPRISES)
SECURITY
EXPERT
SECURITY
EXPERT
SECURITY
EXPERT
OFFER
BOUNTY
ARBITERS
GROUND TRUTH
Figure C: Overview of PolySwarm’s Bounty and Offer lifecycles.
End Users: Enterprise and home users with suspect Artifacts . End Users
participate in the PolySwarm market via Bounties and Offers (more on these
in a moment) and extract timely and accurate malintent classifications
Security Experts (“Experts”): Geographically diverse malware experts
and reverse engineers . Experts dissect the latest suspect Artifacts and
maintain PolySwarm-connected detection engines (“workers”) that determine
malintent . Experts commit to “Assertions”, public statements that reflect the
results of their analysis into the malintent of the Artifact . Those that have
committed an accurate Assertion are rewarded (in NCT) for their efforts .
As far as core participants go, that’s actually it. Simplicity
is good, but this is pretty far from the full PolySwarm story.
Technically, PolySwarm could work with only these two classes of
participants. Realistically, it won’t because most End Users will
prefer to outsource the nitty gritty of interfacing with the PolySwarm
market. Before going further, we must introduce Ambassadors
(another participant) and a certain subclass of these Ambassadors.
p. 08
Version 15
Ambassadors: Companies that make it easy for End Users to benefit from
the PolySwarm market . Ambassadors collect traditional fiat (e .g . subscription
fees) and suspect Artifacts from their clients (End Users) and introduce
Bounties and Offers into the market on their clients’ behalf . It is the
Ambassador’s responsibility to distill the Assertions of various Experts into a
simple malicious or benign Verdict that they deliver to their clients .
A trivial, perhaps naive approach for this distillation might be to
simply average the Assertions provided by Experts. It is unlikely,
however, that such an algorithm would compete favorably against a
Bayesian analysis5, let alone expert human involvement. We expect
existing Antivirus and threat intelligence firms to participate in the
PolySwarm market as early Ambassadors, augmenting their in-house
expertise with PolySwarm-enabled triage of suspect Artifacts.
From the User’s perspective, upgrading to PolySwarm-backed
protection is an easy process: choose a reputable Ambassador and
pay a subscription fee.
An Ambassador’s reputation is based on past Verdict performance
relative to ground truth. Ambassadors are incentivized to make their
Verdicts public due to their inherent desire to build reputation and
attract new clients, as well as take advantage of a discount on
Fees (this process is detailed later). These public Verdicts enable
the creation of an Ambassador “scorecard” that rates real-world
performance against non-synthetic Artifacts with data that is simply
not available in today’s market6.
Arbiters: Top-echelon Ambassadors that are responsible for determining
malintent ground truth . A certain percentage of Ambassadors (in terms of
Fees generated) will be considered “Arbiters” .
During development, PolySwarm will assign Arbitership to existing,
reputable threat intelligence vendors that are willing to maintain
frequent engagement with the PolySwarm team, help identify and
address platform bugs, and help build interest in the ecosystem.
Once PolySwarm 1.0 is ready for launch, these designated Arbiters
will need to maintain top-echelon Ambassador volumes to maintain
Arbiter status.
5 Kantchelian, Alex, et al. “Better malware ground truth: Techniques for
weighting anti-virus vendor labels.” Proceedings of the 8th ACM Workshop
on Artificial Intelligence and Security. ACM, 2015.
6 The closest analog is AV Comparatives (https://www.av-comparatives.
org/).
p. 09
Version 15
Prediction Markets, Arbiters &
Mediated Consensus
In the PolySwarm ecosystem, Security Experts develop micro-engine
workers that compete to quickly & accurately investigate suspect artifacts .
This investigation occurs at machine speed - well before the ground truth
concerning an artifact’s malintent has been established .
This design shares some similarities with prediction markets, i.e.
rewarding past assertions based on accuracy, but differs in two
critical respects:
1. No future data is required to accurately classify an artifact.
The “correct” answer to whether an artifact is malicious can always
be determined with certainty at the time it is made available to
workers. PolySwarm is a skill-based design where nothing is left to
chance7.
2. Determining ground truth will always require expertise. This is
in contrast to the “universal observability” of prediction markets
(including crypto markets such as Augur8) that rely on unskilled
participants to observe and record events as they occur.
Item #1 is PolySwarm’s gambling deterent9.
Item #2 introduces a technical challenge: what is the best way
to incentivize authorities to continually produce ground truth?
PolySwarm’s answer is the Arbiter class and a process we refer to as
Mediated Consensus.
Mediated Consensus is a generic design paradigm that we hope will
find a home in other market design projects. In short, Mediated
Consensus is market design that entrusts a critical task to a subset
of participants. These participants:
1. Are qualified to complete the task (possess expertise).
2. Have their interests aligned with the overall health of the market
(avoiding tragedy of the commons).
7 In theory. In practice, at least two events may introduce uncertainty:
(1) arbiters incorrectly determine ground truth resulting in reward
distribution to wrong experts, (2) the expert’s analysis is accurate
according to ground truth, but the expert disagrees on the boundaries of
malintent - a semantic issue that is larger than PolySwarm. Is adware
malware or simply a potentially unwanted application? Neither of these
variables fundamentally detracts from the skill-based nature of the
PolySwarm ecosystem.
8 https://augur.net/
9 Guessing, betting, gambling or other chance-based wagers on
artifact malintent is bad for everyone and PolySwarm is specifically
designed to be inhospitable to this misuse of the platform.
p. 10
Version 15
We’ve designed Arbiters to satisfy these two constraints. By
definition, the Arbiter class is defined as the most active
Ambassadors in the PolySwarm ecosystem at any given time.
As Ambassadors, Arbiters are contracted by their customers to
distill Security Experts’ assertions into Verdicts. Customers trust
them to possess and exercise automated - and when necessary -
human expertise. These companies already have a vested interest in
maintaining a public record of accuracy (item #1).
As the most active Ambassadors, Arbiters have the most to gain /
lose by an honest / dishonest PolySwarm ecosystem. Their large stake
in the faith of the ecosystem aligns their financial interests with the
health of the market (item #2).
We’ve designed Fees as a supplemental defense against Arbiter abuse
such as Arbiter-Security Expert collusion.
Determining Ground Truth
Most of the technical details of the Arbiter voting process are intentionally
left undefined at this time . The reason for this is simple: the specifics of the
Arbiter decision process are predicated on the specifics of virtually all other
PolySwarm processes - and we expect variations in other processes .
That being said, we expect the following high level design choices to
provide value in this process:
1. Arbiters reach consensus on ground truth via a majority vote.
2. These votes are mined onto the Ethereum chain en masse (many
votes per cycle), saving time and money (Ethereum gas) for the
PolySwarm ecosystem.
3. To encourage participation, Arbiters are rewarded with Fees for
voting on the ground truth of artifacts.
4. Arbiters may abstain from voting on any particular artifact.
Arbiters may choose to do this if, for example, they feel unqualified
to determine the malintent of a particular artifact.
5. As necessary, Arbiter voting privileges are automatically
delegated to additional Ambassadors (in order of volume) to
ensure quorum on all ground truth determinations.
p. 11
Version 15
6. As necessary, we expect participants to challenge Arbiters’
determinations as is done today: with a blog post or technical
paper describing the malintent or benign nature of an artifact that
an authority such as an Arbiter had miscateogrized: an external PR-
driven feedback loop.
The PolySwarm team will iteratively develop the details concerning
Arbiter voting and incentive structures during development. We expect
early Arbiter feedback will be instrumental to this design.
Topics of research include, but are not limited to:
• Minimum Arbiter quorum percentage
• Arbiter deferral procedure (for when quorum cannot be achieved)
• Appropriate Fee reward structure
• Incentive for timely ground truth determination, e.g. only first X
Arbiters to vote receive reward
• Penalty for non-participation, e.g. Arbiters must vote on ground
truth for X% of artifacts in a time window, else their Arbiter status
is revoked
• (As necessary) additional deterrents against perverse incentives
The Instruments
PolySwarm exposes two core instruments to End Users
and Ambassadors10 that increase the efficacy of the threat
intelligence market:
Polyswarm Offers: Requests made directly to reputable Security Experts
for their malintent prediction . PolySwarm provides frictionless access to
thousands of such researchers, enabling traditional information sharing
agreements with non-traditional participants . This interaction happens
inside Raiden11-style Offer Channels (detailed later) . Once Channels
are established, Offers provide millisecond-scale latency for artifact
investigation .
10 Hereafter, End Users and Ambassadors is shortened to
Ambassadors. End Users may act as their own Ambassador.
We expect some large enterprises to participate in this
manner.
11 http://raiden.network/
OFFER
p. 12
Version 15
Polyswarm Bounties: A wild-west style “Wanted” poster with accompanying
Artifact contents (e .g ., Wanted: This Artifact: Malicious or Benign? Reward:
1000 NCT) . Security Experts make a name for themselves (build reputation)
by successfully competing for Bounties . No direct analog exists in today’s
market .
Offers are the closest analogue to today’s on-demand scanning
market and operate with millisecond latency.
Ambassadors issue Offers and Artifacts directly to chosen Security
Experts, optionally under a non-disclosure agreement. Each Expert
chooses whether to accept the Offer based on their confidence in
rendering an accurate Assertion for the provided Artifact. Experts
may choose to decline if they are not confident so as to avoid
adversely affecting their reputation. If the Expert accepts the Offer,
the Expert commits to providing a malintent Assertion in a timely
fashion. Token collection and awards for all instruments are managed
and executed entirely by distributed smart contracts.
Issuing an Offer requires a certain familiarity with Experts
best equipped to dissect the Artifact in question. To ease this
matchmaking, Experts advertise their specialties via listings in the
PolySwarm Worker Registry - PolySwarm’s analog to Ethereum dApp
Registries12 and build public reputation by successfully participating
in similar Bounties.
Bounties are cheaper than Offers and require no upfront familiarity
with specific Experts, but Bounties a프로젝트로 돌아가기