[HN Gopher] Launch HN: Overwatch (YC S22): OSINT platform for cy...
___________________________________________________________________
Launch HN: Overwatch (YC S22): OSINT platform for cyber and fraud
risk
Hey HN! Arjun and Zara here - cofounders of Overwatch
(https://www.overwatchdata.ai), a platform to automate OSINT and
threat intel, turning it into actionable insights. Check out our
clickthrough demo here:
https://app.storylane.io/share/qyayvtamapis. Overwatch began when
we were working with risk and threat intel teams at Google, Stripe,
and government. We experienced the immense challenge every fraud
and cyber threat analyst faces: manually parsing through an ocean
of data to find valuable insights and filter out the noise. This
included using many of the feeds and tools out there that were
often very expensive, noisy, keyword-based, and lacked accurate
entity extraction or advanced query features. Most threat
intelligence tools utilize thousands of keywords and teams of
analysts to manually sift through torrents of alerts. These alerts
are usually individual posts on various platforms across news,
social media, deep and dark web sources that have some matching
keyword. This is full of false positives, requiring many hours to
wade through to figure out what intel matters most to our users,
why, and what they can do next. Overwatch uses an alternative
approach by layering AI agents and NLP techniques, including a
combination of multifarious datasets, cluster analysis, topic
modeling, Retrieval Augmented Language Models (RALM) and domain
knowledgeable agents. This allows us to (1) Filter through OSINT
in real time to identify events and narratives that matter to our
users, and write reports on what they could do about it; (2)
Identify dark web and deep web threats, fraud methods, new tactics,
and compromised accounts, stolen checks, and credentials affecting
our users or their peers; (3) Send an alert any time one a 3rd
party supplier or parts of the tech stack are impacted by a widely
exploited vulnerability, ransomware attack, or breach; and (4)
Track malware and ransomware groups that are actively targeting
your industry including Indicators of Compromise (IOCs). Our
intelligence is actionable because the alert comes with the context
and important details that an analyst needs to make an informed
decision. Being AI-native, we also have a range of chat and data
visualization features to effectively function as an intel co-pilot
or industry expert. Finally, our in-house intelligence analysts and
investigators can assist threat intelligence teams with HUMINT
investigations and darkweb acquisition. Our current customers
include internet platforms, financial institutions, and supply
chain companies. Within a day of one breach, one of our customers
used Overwatch to surface 18,000+ leaked credentials. Another used
us to surface fraudulent checks and learn exactly how threat actors
were targeting their specific product features. Our website says
"Request a demo" but if you want to poke around on a very basic
example of how we're aggregating dark web, deep web, social, and
surface web, log in at https://app.overwatchdata.io/ using these
credentials: username: try_overwatch@overwatchdata.io pw:
HelloHNWorld That login is for an un-personalized feed of cyber
threat intel (breaches, vulnerabilities, ransomed organizations,
and industry updates) that gives you a flavor of not just the kind
of information from which we can collect, but more importantly, how
our technology prioritizes, clusters, and summarizes alerts for
cyber / fraud analysts. Try the chat agent on the left-hand side to
parse through the data. Or sign up for a longer trial and preview
of our email alerts: https://xryl45u9uep.typeform.com/to/pvtZQyS0.
You can also check out our clickthrough demo for dark and deep web
intelligence: https://app.storylane.io/share/qyayvtamapis.
Integration options range from simple dashboard access to our API
for those who want to weave our intelligence directly into other
products. Pricing is dependent on how complex a threat landscape
our users want to monitor and we're still figuring out how to
standardize this but we'll always do our best for the HN community.
Since the platform is AI-powered, it can also be used for news
monitoring, supply chain disruptions, regulatory monitoring, or
social media monitoring. We've had a lot of experience wrangling
text-based feeds and using numerous AI-models (from embeddings,
entity extractors, and LLMs) to filter, categorize, cluster, and
analyze the data into meaning - so let us know if you'd like to
nerd-out or have had any particular challenges. Looking forward to
your feedback and questions! Thanks, HN!
Author : Bisen
Score : 113 points
Date : 2024-06-12 15:32 UTC (7 hours ago)
| ssahoo wrote:
| Congratulations on the launch. I noticed that you guys are SOC2
| type II certified. I wonder how did you achive that so fast?
| Bisen wrote:
| Thank you! It definitely didn't feel quick. We had a 3 month
| audit window and we used Vanta.
| technick wrote:
| Waiting that 12 months to really demonstrate you have a
| working security program with efficient controls really pays
| off. It's something I look for when doing vendor reviews and
| I assume others do the same.
| Bisen wrote:
| That makes sense, we're going through our annual renewal
| now. It's a great experience to harden and test systems.
| patchorang wrote:
| Sort of off topic question. But how would you get into the type
| of work that uses this tool? I've always thought this type of
| work would be interesting, but I have no ideas where to start.
| What are the job titles? Fraud Analyst? Thanks!
| zara2 wrote:
| I love it! We've seen the job title change a bit depending on
| the sector but can be under fraud analyst, threat intel
| analyst, risk analyst, financial crime analyst, sometimes trust
| and safety teams.
| PenguinCoder wrote:
| Titles vary greatly but the general domain is information
| security and cyber security. Here is a good primary of basic
| qualifications and duties for a given role (Generic, not tuned
| to a specific company) - https://www.cyberseek.org/pathway.html
|
| This type of information (OSINT of vulns/cves, proof of
| concepts) is useful for the Blue team side of defending against
| attackers. With easy to access information in a timely manner,
| the defenders can proactively put roadblocks and alerts into
| place for vulnerabilities as opposes to AFTER they are
| popped/hacked by such.
|
| Prevention is ideal; detection, a must.
| Tao3300 wrote:
| There's no getting into anything you haven't already been doing
| 6+ years right now.
| candiddevmike wrote:
| What is your detection hit/miss rate? What happens when you miss
| something?
|
| Seems like this is going to become a cat and mouse game similar
| to evading AV.
| zara2 wrote:
| Great point! From some of our case studies we see users catch
| 25% more 3 days earlier than other solutions.
|
| To your point catching every threat or every alert especially
| on darkweb is always a cat and mouse game. Our idea is a
| prioritization problem - how do you mitigate the biggest risks
| quickly.
|
| The existing OSINT tools we used are keyword search based /
| pretty noisy so we've been focusing on the idea that given
| there's no way analysts can find or triage every alert, how do
| you catch the biggest stuff. We do a few things from AI
| crawlers to continue to expand data collections to AI
| categorization, clustering, data extraction etc to make it
| easier to track the cover the most ground.
| candiddevmike wrote:
| I find it interesting that you didn't answer my question at
| all, tbh.
| sbarre wrote:
| I doubt they provide any kind of guarantee that they'll
| catch everything. That's not what a tool like this is used
| for (i.e. it's not a security tool or a network monitoring
| service).
|
| OSINT is about gathering data from public or semi-public
| places.
|
| There are plenty of private dark web forums that these
| kinds of tools probably don't have visibility into, but the
| bigger public well-known ones are where you are most likely
| to see high-profile breaches being sold (since there's a
| wider audience and actors want to sell data quickly before
| it devalues), so it's certainly better than nothing...
| candiddevmike wrote:
| I never asked for a guarantee, that's not feasible, I was
| looking to see if in-house analysts or engineers can
| review the data they're using or lower confidence results
| as a second set of eyes.
| sbarre wrote:
| Ah well, your question wasn't clear. But your
| clarification helps!
| sam0x17 wrote:
| it probably varies widely by use-case and customer
| edm0nd wrote:
| Dodging questions and proving a non-answer answer is always
| the sales and c-level route when they just dont want to
| answer you (or know the answer).
| skilled wrote:
| This looks great, I am seeing a lot of potential use cases here.
| It also pools together a lot of the stuff that goes unreported in
| the news.
|
| Would this be a service you would ever offer to regular
| researchers?
| zara2 wrote:
| Definitely! Would love to chat more. A lot of our users want
| more customized monitoring / agent versions of this, and our
| dashboards are pretty easy to customize.
|
| Opening up certain components of the platform is something we
| are definitely looking into and passionate about.
| chicagojoe wrote:
| FWIW, clicking around, there are some odd display issues in the
| "References"
| (h<b>ttp</b>s://attack.<b>mitre</b>.org/techniques/<b>T1486</b>)
|
| It looks like you're embedding data from Twitter - are you paying
| for decahose/enterprise access or just paying for a low volume of
| high value tweets (i.e. I'm seeing many from FalconFeedsio,
| DailyDarkWeb)
| Bisen wrote:
| I can jump in on the data source question - we can track
| specific accounts and keywords on Twitter, we aren't paying for
| the full firehose yet. We also track those original ransomware
| and dark web sites and blogs being referenced and just figuring
| out how best to cluster them all into the same event. Thanks
| for checking it out!
| martinbaun wrote:
| Congrats, look very good!
|
| I would hesitate with the name though. Overwatch is also a game
| series from Blizzard.
|
| And Blizzard is known to be a little sue-addicted.
| zdw wrote:
| AI tool with the same name as a game where a robot uprising is
| one of the main backstories seems like it could be a bit
| problematic...
| jrockway wrote:
| Isn't Dota 2's anti-cheat system called Overwatch? If
| Blizzard didn't go after that, they aren't going to care
| about this.
| floundy wrote:
| Valve used the name Overwatch since 2013 (before Blizzard)
| for their player-run jury system for cheaters.
| Bisen wrote:
| Haha! It's also an SEO nightmare. The term Overwatch is a
| force protection tactic, where you have one team providing
| extra assistance and cover to the main operation. If only we
| had played the game first before naming ourselves!
| ycombinatorics wrote:
| should have at least named it Overwatch 3 to get some SEO
| benefits
| jacques_chester wrote:
| Trademarks are namespaced by subject matter.
| kube-system wrote:
| Yep, Blizzard's trademarks are for:
|
| > Computer game software, [ computer game discs, ]
| downloadable computer game programs, computer game software
| downloadable from a global computer network, electronic games
| software for wireless devices, interactive multimedia
| computer game programs, mousepads, computer mouses, headsets
| for use with computers, computer keyboards
|
| > Printed matter, namely, [ computer game strategy guides, ]
| comic books, graphic novels, novels, art books, calendars,
| posters, [ notebooks, ] and stickers
|
| > Entertainment services, namely, providing on-line computer
| games; Providing computer games that are accessed via a
| global computer network
|
| > Clothing, and headgear, namely, caps, hats, hooded
| sweatshirts, jackets, sweaters, and T-shirts
|
| > Toys, games, and playthings, namely, action figures,
| collectible toy figures, dolls, plush toys, and vinyl toy
| figures
|
| https://tsdr.uspto.gov/#caseNumber=86239314&caseSearchType=U.
| ..
|
| https://tsdr.uspto.gov/#caseNumber=86434530&caseSearchType=U.
| ..
|
| https://tsdr.uspto.gov/#caseNumber=86239318&caseSearchType=U.
| ..
|
| https://tsdr.uspto.gov/#caseNumber=86980853&caseSearchType=U.
| ..
|
| A search for "overwatch" currently matches 191 registered
| trademarks in the US.
| martinbaun wrote:
| thanks Kube said it lot better than I did here. :)
| kube-system wrote:
| I'm saying they _shouldn 't_ be worried about being sued
| by Blizzard. This software is not a video game or related
| merchandise.
|
| Trademarks do not cover all uses of a word -- they only
| cover the use of a word in relation to a particular field
| of commerce. This is, likewise, why your grocery store
| can sell apples without being afraid of the equally
| litigious Apple Inc.
| Tao3300 wrote:
| What about WWF? I always thought you'd have to be a
| complete imbecile or willfully dishonest to confuse the
| World Wrestling Federation with the World Wildlife Fund,
| yet here we are.
| GloriousKoji wrote:
| But they still would need to watch our for the even more
| litigious Apple Corp.
| whoopdedo wrote:
| The number of times I've seen someone market a product called
| "The Matrix" since 1999 would tell me trademarks like this
| aren't a big deal.
| zelias wrote:
| Maybe split the difference by naming Sombra the CTO
| keepamovin wrote:
| I disagree - I think the name is great.
|
| Consider Venn diagrams: the audience for these two homynymous
| products has small overlap. Further moderated when you consider
| term stickiness to respective meanings is only high for a small
| fraction of that audience.
|
| In other words, most people aware of both can cope with a
| mutual name. And some may even think it's cool. Each name
| enhancing the other through association and analogy.
|
| Overwatch is definitely the right choice. Consider the 'OG'
| meaning originates from military terminology. In this context,
| "overwatch" refers to a tactical position where one unit
| provides covering fire and surveillance for another unit as it
| moves forward or performs an action. The overwatch position is
| typically elevated or strategically placed to have a clear view
| of the battlefield, allowing the overwatching unit to detect
| threats and engage enemies to protect the advancing or exposed
| units.
|
| This concept ensures that the moving or vulnerable unit can
| operate with reduced risk, as the overwatching unit can
| neutralize potential dangers and provide critical information
| about the surroundings. The practice of overwatch is a
| fundamental tactic in military operations, emphasizing
| teamwork, communication, and strategic positioning.
| bogtog wrote:
| I'm not sure if you are familiar with Overwatch the game, but
| it is a very famous massive game. My gut is that a majority
| (if not a huge majority) of people in tech aged 20-40 will be
| familiar with Overwatch, the game. This will just make
| following a conversation about the product unnecessarily
| difficult. The existing mental and SEO associations are
| radically strong, regardless of theoretical reasons for why
| "Overwatch" is a good name for the present product
| martinbaun wrote:
| not saying the name is great. I am just saying maybe you
| don't want to get sued by a huge multi-billion dollar giant
| who likes to sue.
| intended wrote:
| Its a great name, it is also a trademark. I dont know enough
| IP law to know if this is an issue.
| https://trademarks.justia.com/862/39/overwatch-86239314.html
|
| Try it and find out?
| ygjb wrote:
| Meh. That is just the Blizzard trademark; there are several
| other Overwatch trademarks that are military and security
| related that are probably a more likely overlap.
|
| Definitely something to consider, but more in the "send a
| note to a trademark lawyer and marketing folks" than
| "rename our entire product and brand".
| marcus0x62 wrote:
| It is also, and more to the point of being a real trademark
| issue, the name of a managed threat hunting service from
| Crowdstrike.
| technick wrote:
| Doesn't the government have a program named overwatch as well?
| Something in the air force cyber transport division...
| waihtis wrote:
| Pretty cool, like the fact you tie all types of feeds (threats,
| vulns etc) together into a single view. How are you guys
| different from the plethora of other TI platform vendors out
| there?
| Bisen wrote:
| In 3 important ways besides price: -Personalization: The
| platform can be fully customized to your interests, e.g. 3rd
| party vendors, tech stack, products, peers or industry. It's
| like having your personalized threat intel org that cuts
| through the noise. Each of the alerts are ranked and tailored
| to your interests. -Customization: The platform can be used for
| a range of use cases, with agents undertaking tasks ranging
| from finding and extracting check, loan, and credit card fraud,
| risky narratives about your brand, through to breaches or
| emerging ransomware groups targeting your tech stack or
| vendors. Those agents can even identify breaking events that
| could be near your assets. -So what and now what: Each report
| provides finished intel, whether finding or extracting relevant
| indicators of compromise, background on the threat actor and
| victim, compromised credentials, or compromised credit cards
| and checks. We can even automate workflows through integrations
| or creating cases/escalations for specific teams.
| abtinf wrote:
| How is this different from Interpres?
| Bisen wrote:
| In 3 important ways besides price:
|
| -Personalization: The platform can be fully customized to your
| interests, e.g. 3rd party vendors, tech stack, products, peers
| or industry. It's like having your personalized threat intel
| org that cuts through the noise. Each of the alerts are ranked
| and tailored to your interests.
|
| -Customization: The platform can be used for a range of use
| cases, with agents undertaking tasks ranging from finding and
| extracting check, loan, and credit card fraud, risky narratives
| about your brand, through to breaches or emerging ransomware
| groups targeting your tech stack or vendors. Those agents can
| even identify breaking events that could be near your assets.
|
| -So what and now what: Each report provides finished intel,
| whether finding or extracting relevant indicators of
| compromise, background on the threat actor and victim,
| compromised credentials, or compromised credit cards and
| checks. We're training our agents to be even more specific with
| answers e.g. "what IOCs relate to malware groups most active in
| the airline industry". We can even automate workflows through
| integrations or creating cases/escalations for specific teams.
| thecleaner wrote:
| Wow. Thats a really fresh concept and an important need. Congrats
| on the launch.
| devmor wrote:
| Using RAG is definitely a relief factor after reading that you're
| using AI and NLP for aggregate analysis, but I'm curious how much
| manual review this actually saves?
|
| Since the model summaries would still need to be validated
| against the source results manually, your business' actual
| viability as a product hinges on whether customers perceive a
| significant time savings in the data provided via these channels
| over historical aggregation methods (like keyword analysis that
| you mentioned) and level of false positives.
|
| What do you measure as the largest impact here? Is there a large
| time savings, is it additional discovery from blindspots that
| other methods don't cover? Both? Are there additional benefits
| you see to this model beyond automation and expanded discovery?
| Bisen wrote:
| Some of our customers said they spend around 3hrs a day on
| navigating new vulnerabilities alone. Step 1: wading through
| info from some easy and some hard to access sources; Step 2:
| trying to bring together all the most relevant information.
| E.g. just detecting your payroll provider got ransomed is just
| one step, then you have to research the group, any indicators
| that you might also be infected etc. then step 3: what do I do
| next? e.g. adding relevant hashes to virus total.
|
| We not just help with relevant detection, but also automate
| some of the next two steps as well. Bringing the total weekly
| time saved down to a few minutes a day.
| Bisen wrote:
| The discover is another important value add - AI Agents can
| monitor beyond human scale across more data sets to do the
| initial triaging for you.
| artembugara wrote:
| Arjun and Zara are amazing! They're our batch and group office
| hours mates from YC S22.
|
| We (https://www.newscatcherapi.com/) also serve the same use case
| but only for the news analysis part. And we don't really have a
| UI: it's all data accessible via an API.
|
| I see a lot of questions here about comparing Overwatch to other
| OSINT tools. The ability to customize/personalize is a huge
| difference.
|
| In my experience, clients with the most expensive problems are
| super underserved because there is no "Palantir-like" solutions.
|
| Don't get me wrong: you don't have to do consulting -- just tweak
| the onboarding/set up. Making bespoke solution for the companies
| with the biggest problems is a great way to get into the market.
| And it can work on the huge scale. E.g. Palantir.
|
| An example from what we have as a very typical situation at
| NewsCatcher: a big bank is absolutely blown away because we
| actually can find news about private companies that they need to
| track with minimum false positives. And all we have to do is to
| tweak a bit our entity disambiguation module to work with the
| data points that the bank actually has.
| ThinkBeat wrote:
| This feels a bit like a turbo RSS reader that plows through some
| easy and some difficult to access information and actively
| selects and targets it to subscribers?
| Bisen wrote:
| That's definitely the vibe of the example we're presenting. But
| for specific customers, the agent can cluster and bring
| additional relevant context to each of the 'events', and even
| recommend actions / automate certain actions e.g. if you detect
| a compromised account, send it team X.
| scrollaway wrote:
| Would love to chat with one of the founders. Can you send me an
| email? (profile)
| dang wrote:
| Just an offtopic heads-up that emails in HN profiles aren't
| visible to other users, only to admins.
|
| If anyone wants to share an email address with other users, it
| needs to go in the About box.
| adi_lancey wrote:
| Great team, great launch!
| orliesaurus wrote:
| I hope it's not as bad as Overwatch 2 /s
| edm0nd wrote:
| What is the pricing to monitor per each keyword?
|
| I know platforms like Flare are cool but when you need to monitor
| hundreds or even thousands of corporate keywords, domains, and
| assets, it becomes cheaper for CTI to just write the tools
| themselves.
|
| What does your platform look like in regards to this and pricing?
|
| For example, your pricing for monitoring 100 keywords and pricing
| for monitoring 500 keywords.
|
| 200k unique telegram channels is an interesting stat.
|
| Each Telegram account (if paid Premium account) can only be in 1k
| channels and groups max. To monitor 200k unique channels/groups,
| you have a network of at least 200 paid Telegram accounts
| continuously monitoring? Are you using Pyrogram or Telethon for
| this? Are these accounts owned by you (Overwatch) or are you just
| using a bunch of 3rd party Telegram intel feeds?
| Bisen wrote:
| We totally hear you and that's why we don't really charge by
| keyword but instead look at how many agents we need to deploy/
| use cases we build towards. A 1000+ assets is the norm for our
| users. Would you be interested in connecting on a call to
| better understand the use cases and tell you more about how it
| works?
| airstrike wrote:
| How is this different from RecordedFuture? Other than the obvious
| RAG capabilities
| redman25 wrote:
| How do you compare to incumbents like Blumira? What does your
| mitre coverage look like?
| guyseneca wrote:
| Seems like alerting in threat intel is getting disrupted by AI -
| Cool.
| jonnyparris wrote:
| Congrats on the launch! As design feedback, the demos don't seem
| to pass the "squint test" for intuitively surfacing the most
| important information / actions on the screen. Maybe a more
| specific walkthrough of a redacted / hypothetical scenario that's
| focused more on the user's decision-making process & actions
| instead of the kitchen sink of product features would better
| illustrate how/why things are laid out as they are currently.
| ssahoo wrote:
| Also storyline type demo sucks. A simple video that I can
| rewind will be preferred.
| baxtr wrote:
| Thanks. Who is the target user for this kind of tool? A CISO team
| member?
| Bisen wrote:
| Our current users are CISO's security ops team, threat intel
| team, blue team, fraud strategy or fraud intel team. Hope that
| helps!
| sncsy wrote:
| I worked with Arjun in trust and safety / risk at both Google and
| Stripe. He's not only an expert in the space, but is incredibly
| users-first. If you're looking for a product like this and want a
| great partner, Arjun and team are it!
___________________________________________________________________
(page generated 2024-06-12 23:00 UTC)