Tag Archives: Opinions

Assessment of DOGE and Musk’s Actions for Government Efficiency

от Божидар Божанов

лиценз CC BY

The Trump administration began with a commitment to optimizing the efficiency of the government, introducing the so-called DOGE (Department of Government Efficiency) under Elon Musk. On the first day, I wrote that I would follow DOGE’s actions in optimizing the administration with interest. This is a task that we also face in my country, with electronic governance (of which I was the first minister a few years ago) being a tool for achieving it. On the very day of the executive order establishing DOGE, I wrote to colleagues that “the risky part is that every structure will have to provide them with all non-classified documents” and that they would likely be looking for “skeletons in the closet.”

This risk, stemming from the executive order, quickly materialized due to the actions of DOGE employees (or USDS, which is the U.S. government’s structure for IT services and administrative modernization, serving as DOGE’s main tool). The employees, who are 20-25-year-old IT specialists, gained access to the systems and databases of several key government agencies, even connecting external hard drives. This understandably sparked serious dissatisfaction.

I do not dispute the right of specialized units to access data and documents across the entire government structure. Nor do I deny the right to close down agencies or terminate funding — these are political decisions (subject to judicial oversight) that any administration has the authority to make (and anyone has the right to disagree and challenge them in court). Moreover, fraud and inefficiencies in spending certainly exist, and an external review of the data and systems can identify and eliminate them.

However, for the implementation of these political decisions to be legitimate, it must follow established rules. It appears that Musk and his team operate under the assumption that they are above the rules — that they can take over the information systems of key government agencies through “top-down orders.” These actions should be carried out properly, especially considering information security and data protection, which are at risk from such ad-hoc measures.

I will give a few examples from my time as a minister and earlier as an advisor, fully aware that the scale is different, but there are still direct parallels.

The first example is the SEBRA system (the Ministry of Finance’s budget payment system). Musk is essentially doing the same thing — extracting and potentially publishing all state payments. However, we did it properly — together with the Minister of Finance, we submitted a proposal to the Council of Ministers, designating payment data as a priority data set for publishing in compliance with the Access to Public Information Act. This decision mandated the Ministry of Finance to provide the data to the Ministry of e-Government in a specific format so that it could be processed, anonymized, and published. The Ministry of Finance assigned its contractor to extract the data and securely transfer it to the Ministry of e-Government, where the processing and anonymization code was published in an open-source repository. At the same time, a working group revised a regulation so that the system would automatically publish such data. If we had done it Musk’s way, I would have simply sent one of my advisors and told them, “Plug in your laptop and get access to the database.” And that would have been wrong.

In 2016, we opened up data from the Commercial Register and the Public Procurement Register. At the time, I was an advisor, but the data disclosure process involved official correspondence, and I never had direct access to the database — experts from the respective institutions wrote queries to the database, and I received a test database without real data to work on queries for extracting information. We also published the code for the anonymization tool used for the Commercial Register. This was all done based on formal communication and a valid legal basis under the Access to Public Information Act.

As a minister (after all, officially elected by parliament, not an unknown staffer), I requested to review systems, but someone else always had to show them to me while I only directed what should be extracted (even though I could have found it myself much faster). In these cases, the approach was always, “Sit here and show me,” or “Based on [legal grounds], please provide me with the following data,” rather than “Give me access to the database and leave.”

Yes, there is a risk of refusal or sabotage of these efforts, but there are other measures for that—those who sabotage can be reassigned or even dismissed if they do not comply with a lawful order. However, Musk is doing the opposite—his people (who have not been properly vetted) are acting unlawfully and are being stopped by the courts. In some places, internal teams correctly identified these actions as an “insider threat.” In principle, every access to databases should leave a trace, and access for usual personnel must be controlled because data misuse is not only a risk from external actors but also from internal ones.

The issue of access to institutional data is extremely serious and cannot be resolved with a blanket statement in a presidential executive order. It is serious because a modern state largely depends on its registers and databases. Their security and integrity are fundamental to many policies and their success or failure. “Seizing control” of registers and databases is a key instrument of power with significant negative potential.

Beyond this specific case, if DOGE had acted “properly,” it would have taken a few more months, but its actions would have had more legitimacy. However, in line with other actions and statements by the U.S. government, it seems that they are more interested in breaking things and disregarding the rules.

Government efficiency is important, and reducing unnecessary expenses is a valid political priority. But while it can be a legitimate goal, it can also serve as a cover for less honorable intentions. And although I come from the startup world, I do not agree with the approach that a government and administration can be run like a startup. In startups, the goal is to move fast, even at the risk of breaking things. In the public sector, it is more important not to break things, because lives and livelihoods depend on it, which is why processes take longer. And when you act by breaking the rules under the justification of political expediency, you do not build trust — you destroy it.

The post Assessment of DOGE and Musk’s Actions for Government Efficiency appeared first on Bozho's tech blog.

Is Ransomware Protection Working?

от Божидар Божанов

лиценз CC BY

Recently an organization I’m familiar with suffered a ransomware attack. A LockBit ransomware infected a workstation and spread to a few servers. The attack was contained thanks to a quick incident response, but there was one important thing: the organization had a top-notch EDR installed. It’s in the Gartner’s leader quadrant for endpoint protection platforms. And it didn’t stop it. I won’t mention names, because the point here is not to attack a particular brand, but the fact that the attackers created a different build of the LockBit codebase meant that the signature-based detection didn’t work and so the files on the endpoints were encrypted. Not only that – attackers downloaded port scanners and executed password spraying on the network, in order to pivot to the servers, without much trouble caused by the EDR.

Ransomware is the most widespread type of attack, we have a security industry that’s trying to cope with those attacks for a long time, and in 2024 it’s sufficient to just rebuild the ransomware from source in order to not get caught. That’s a problem. To add insult to injury, the malware executable was called lockbit.exe_.

I think we need to think of better ways to tackle ransomware. I’m not an antivirus/EDR in-depth expert, but I have been thinking the following: why doesn’t the operating system have a built-in anti-ransomware functionality? The malware can stop endpoint agents, but it can’t stop the OS (well, it can stop OS services or modify OS dlls, I agree). I was thinking of native NTFS listeners to detect ransomware behavior – very fast enumeration and modification of files. Then I read this article titled “Microsoft Can Fix Ransomware Tomorrow” which echoed my thoughts. It’s not that simple as the title implies, and the article goes to argue that it’s hard, but something has to be done.

This may seriously impact the EDR market – if the top threat – ransomware – is automatically contained by built-in functionality, why purchase an EDR? Obviously, there are many other reasons to do so, but every budget cut will start from the EDR. Ransomware has been a successful driver for more investment in security. To prevent putting too much functionality in the OS, EDRs can tap into the detection capability of the OS to take further actions.

But ransomware works in a very simple way. Detecting mass-encryption of files is easy. Rate-limiting it is reasonable. You may be aware, but CPUs have AES-specific instructions that improve the performance of AES encryption – the fact that these instructions are used may be a further indicator of an ongoing ransomware attack.

I think we have to tap into the lower levels of our software stack – file system, OS, CPU – in order to tackle ransomware. It’s clear that omnipotent endpoint agents don’t exist, no matter how much “AI magic” you sprinkle ontop of them in the data sheet (I remember once asking another EDR vendor on a conference how exactly their AI is working. My impression from the answer was: it’s rules).

As I said, I’m no AV/EDR expert, and I expect comments like “But we are already doing that”. I’m sure APIs like this one are utilized by AV/EDRs, but they may be too slow or too heavy to be used. And it may mean that this APIs can be optimized for the ransomware-detection usecase. I don’t have a ready answer (otherwise I’d be an EDR vendor), but I’d welcome a serious discussion on that. We can’t be in a situation where purchasing an expensive security tool doesn’t reliably solve the most prominent threat – “off-the-shelf” ransomware.

The post Is Ransomware Protection Working? appeared first on Bozho's tech blog.

The xz Backdoor Should Not Happen Again

от Божидар Божанов

лиценз CC BY

A few days ago a significant supply chain attack attempt was accidentally revealed – the xz utiliy was compromised, likely by a nation state, in order to plant a backdoor which allows sniffing on encrypted traffic.

The xz library is a building block of many other packages and is basically ubiquitous. A famous XKCD strip describes the situation graphically:

This means that if it wasn’t accidentally discovered due to worsened performance, we would eventually have a carefully planted backdoor on practically every Linux server out there. This is a major issue and even though open source security is better than closed source security, even if just by allowing backdoors to be discovered by anyone, we need to address such nation state attempts of planting backdoors.

I propose two complementary measures:

Public funding for open source – the EU and the US need to create a structured, not overly bureaucratic process to fund the maintenance of core open source projects (like xz). Germany has done a good job in setting up its Sovereign tech fund, but we need broader instruments that make sure there is no open source abandonware on which many other projects depend. Currently large corporations often fund the development of open source, but xz is an example that the little building blocks may fall through the cracks. Open source funding can also be directed at systematic security analysis of open source projects (like the one in point 2, but not limited the security services).
Analyzing high-risk project – security services and other public and private organizations need to first pinpoint high-risk projects (ones that if compromised, cause a huge risk that trickles down to the whole ecosystem), rank projects based on risk, and then analyze no just source code, but also maintenance activities, maintainer recruitment and churn, commit patterns and so on. In hindsight, the xz backdoor could have been caught by monitoring such metadata and the suspicious activities by the “hacker”. We, of course, need (open source) tools to do these analysis, but also highly-skilled people in the security services of larger countries.

Overall, we can and should learn lessons and take measures based on this incident. Because the next one might not cause noticeable performance degradation and get into actual production, which will be devastating.

The post The xz Backdoor Should Not Happen Again appeared first on Bozho's tech blog.

Why Facebook’s Lack of Customer Support Is a Problem

от Божидар Божанов

лиценз CC BY

Facebook is arguably the biggest social network. The network effect makes it hard for people to leave Facebook, and so many businesses, celebrities, institutions, politicians rely on it for reaching out to their customers/fans/citizens/voters.

Yet, at least in my part of the world, the customer support of Facebook is practically non-existent. Because I’m a member of parliament and former minister that had handled disinformation and relations with Meta, many people turn to me for their Facebook woes. And they are almost never resolved.

A few examples: a deep fake of the Bulgarian prime minister was circulating on Facebook for several days, after two institutions submitted official take-down notices. Profiles of fellow members of parliament were blocked/hacked. None of their support requests succeeded and their profiles remained blocked for months. A fellow member of parliament with paid subscription could not change his cover photo during an election campaign for mayor, and Facebook’s support stopped answering. Facebook bulk-deleted our candidate pages after one election campaign (after it has been taking ad money), and its support did not respond adequately (pages remained deleted). One colleague’s ad account was hacked and a malicious actor used his credit card to promote ads. He was unable to remove the intruder and Facebook’s support didn’t manage to do it either, so my colleagues had to remove the credit card. When I became a minister, my request for a blue checkmark was initially rejected and the official support channel didn’t answer. And in all of those cases support was requested in English, so it’s not about language-specific limitations.

I’m sure anyone using Facebook for business has similar experiences. In a nutshell, support is useless, even if you are paying customer or advertiser. And clearly there is no market pressure to change that.

The European Union recently introduced the Digital Services Act which at least pushes forward a long-time proposal of mine for appeals and independent arbitration for decisions that block access. I don’t know if that’s working already, but at least it’s a step.

So why is that a problem? Facebook argues it is not a ‘natural monopoly’, and I’ll agree with that to an extent – it faces competition from different types of social networks. But its scale and the network effect means it is not just a regular market player – it is (as the digital services act puts it) – a very large online platform that has gained a broad influence and therefore needs to be required to bear extra responsibility. The ability for some entity with 4 million users in a country of 7 million to arbitrarily ban members of parliament or candidates for mayors, or to choose (because of inefficiency) to leave a deep fake of a prime minister up for days, is a systemic risk. It’s a systemic risk to leave a business to be reliant on the whims and inefficiencies of the nearly non-existent customer support.

If a company can’t get customer support sorted, market forces usually push it out of the market. But because of the network effect (and its policy of acquiring some potential competitors), this hasn’t been the case. And if one of the most highly-valued companies on earth can’t have a decent support process, regulators should step up and set standards.

The post Why Facebook’s Lack of Customer Support Is a Problem appeared first on Bozho's tech blog.

Another Attack Vector For SMS Interception

от Божидар Божанов

лиценз CC BY

SMS codes for 2FAs have been discussed for a long time, and everyone knowledgeable in security knows they are not secure. What’s more – you should remove your phone number from sensitive services like Gmail, because if an attacker can fallback to SMS, the account is compromised.

Many authors have discussed how insecure SMS is, including Brian Krebbs, citing an example of NetNumber ID abuse, in addition to SIM Swap attacks and SS7 vulnerabilities.

Another aspect that I recently thought about is again related to intermediaries. Bulk SMS resellers integrate with various telecoms around the globe and accept outgoing SMS by API calls, which they then forward to a telecom in the relevant country. Global companies that send a lot of SMS try to look for cheap bulk deals (you are probably aware that Twitter/X recently decided to charge for SMS 2FA, because it was incurring high costs). These services are sometimes called a2p (application to person).

This means that intermediaries receive the 2FA code before it reaches the subscriber. A malicious insider or an attacker that compromises those intermediaries can thus have access to 2FA codes before they reach the subscriber.

I don’t know if this attack vector has been used, but it is a valid attack – if an attacker knows the intermediaries that a given service is using, they can either try to compromise the systems of the intermediaries, or gain access through a compromised insider. In either scenario, the victim’s 2FA code will be accessible to the attacker in real time.

This just reinforces the rule of thumb – don’t rely on SMS for two-factor authentication.

The post Another Attack Vector For SMS Interception appeared first on Bozho's tech blog.

eIDAS 2.0, QWACs And The Security Of The Web

от Божидар Божанов

лиценз CC BY

Tension has been high in the past months regarding a proposed change to the European eIDAS regulation which defines trust services, digital identity, and the so-called QWACs – qualified website authentication certificates. The proposal aims at making sure that EU-issued certificates are recognized by browsers. Here’s a summary from Scott Helme, and a discussion with Troy Hunt, and another good post by Eric Rescorla, former Firefox CTO so I’ll skip the intro.

Objections

Early in the process, Mozilla issued a position paper that raises some issues with the proposal. One of them is that what the EU suggests is basically an Extended Validation certificate – something that we had in the past (remember the big green address bars?), and which we have abandoned some time ago, and for a reason – multiple studies found that they do not bring any benefits. The EU says “QWACs (EVs) give the user more trust because they know which legal entity is behind a given website”. And the expert community says “well, in which scenario is that useful, and what about faking it – opening an entity with the same name in a different jurisdiction?”.

Later in the process, an additional limitation was added for browser vendors – that they cannot mandate additional security requirements than those specified by the EU standards body – ETSI. This is, to me, counterintuitive policy-wise, because in general, you set minimum requirements in regulations, not maximum. Of course, this prevents browser vendors from having arbitrary requirements. Not that they’ve had such requirements per se, but for example their CA inclusion page says “Mozilla is under no obligation to explain the reasoning behind any inclusion decision.” For me, this is not an acceptable process for something as important.

Mozilla (and various experts) also note, correctly, that if a CA gets compromised, this affects the entire world – the traffic to any website can be sniffed through man-in-the-middle attacks. And this has happened before. The Electronic Frontier Foundation, a respected digital rights organization, also objected to the approach.

Then Mozilla launched a campaign website against the amendment, which has the wrong tone and has some gross oversimplifications and factually incorrect statements (for example, it’s not true that QTSPs are not independently vetted). Then the European Signature Dialog (basically, an association of EU CAs, called QTSPs – qualified trust service providers), responded to it in a similarly inappropriate way. It said “Mozilla is generally perceived as a Google satellite, paving the way for Google to push through its own commercial interests” (which is false, but let’s not go into that).

The statements that QWACs are better against phishing, is arguably not true, even if you consult the paper that the ESD linked. It says: “Our analysis shows that it is generally impossible to differentiate between benign sites and phishing sites based on the content of their certificates alone. However, we present empirical evidence that current phishing websites for popular targets do typically not replicate the issuer and subject information”. So the fact that phishing sites don’t bother using EVs is somehow a reasons that EVs(QWACs) help against phishing? I’m disappointed by this ESD piece – they know better. Mozilla also knows better, as this negative campaign website introduces a tone that’s not constructive.

Insufficient assessment

What becomes apparent from the impact assessment study, another study, and the subsequent impact assessment is that there have been efforts to agree with browser vendors on including EU issued certificates without the CAs having to go through the root program process of the browsers, which they have refused.

I think this is not a good impact assessment. It does not assess impact. It doesn’t try to find out what will happen once this is passed, nor it tries to compare root programs with current ETSI standards to find the gaps. Neither the initial study, nor the impact assessment review the security aspects of the change.

For example, due the current usage patterns of QWACs for internal API-based communication between EU institutions, QWACs have sometimes been issued to private addresses (e.g. 192.186.xx.xx). Once they become automatically approved by the browsers, a security risk arises – what if I have a trusted certificate for your local router IP?

Also, it’s not a good process to include additional limitations in the trialogue, which is an informal process between the EU parliament, commission and council. I, as a Bulgarian member of parliament, requested from our government the drafts from the trialogue, and I was not granted access (due to EU rules). This is unacceptable as a legislative process, which should be fully transparent.

I have criticized this process and insufficient impact assessments before – for the copyright directive introduction of a requirement for automated content takedown, and for the introduction of mandatory fingerprints in ID cards.. There just doesn’t seem to be enough technical justification for regulations that have a very significant technical impact – not just in the EU, but in the world (as browsers have a global trusted CA list, not a EU one).

Technical or political debate?

The debate, as it seems, has conflated two separate issues – the technical and the political one. What Mozilla (and I presume other browser vendors) are implying is that they are responsible for the security in their browsers and they should be able to enforce security rules, while the EU is saying – private US entities (for-profit or non-profit) cannot have full control over who gets trusted and who doesn’t. Both are valid arguments. The EU seems to be pursuing a digital sovereignty agenda here, which, strategically, is a good idea.

But the question is whether that’s the best approach, and if not – how to improve it.

Some data and an anecdote to further illustrate the status quo. The Certinomis French QTSP (CA) has been distrusted by Mozilla a while ago. It is, however, on the EU trusted list. With the changes, Mozilla and others should re-trust it. The concerns raised by Mozilla seem legitimate, and so the fact that EU conformity assessment bodies do regular audits may not be sufficient for the purposes of web security (but this assumption needs to also be critically assessed).

Currently, there are 146 root/subordinate CA certificates listed for QWAC QTSPs. Of those 146, 115 are included in one or more root trust stores, and between 57 and 80 are missing from one or more. It’s far from an ideal picture, but these numbers should have been in the impact assessment and the problem statement in the first place. So that the legislators can identify the reasons for not including a CA in one or more root program. Is it a technical shortcoming of the CA, or it it a vendor discretion? Certainly, there doesn’t seem to be a ban on EU CAs/QTSPs by browser vendors.

So, essentially, the political results here is that EU CAs will get a fast-track into trust stores. I’m sure other jurisdictions will try to pass similar legislation, which will complicate the scene even further. Some of them will not be as democratic as the EU. And if a browser vendor thinks some CA is not trustworthy by their standards, they may come up with very clever workarounds of the regulation.

The first one – ignoring it, as there are no fines. But they can also introduce different paddock colors for different cases – e.g. a QWAC from a browser-approved CA gets a green paddock, a QWAC that has not passed the browser root program gets a yellow paddock. Compared to the current grey one, the yellow color may be perceived as less trustworthy. And then we’ll have to argue whether yellow is a clear enough indication and whether it shows trust or not so much. Time and time again I have stated that you can’t really regulate exact features and UI.

A technical solution to the political question?

I’ve heard many times that technical solutions to political problems are wrong. And I’ve seen many cases where, if you delve into the right level of detail, there is a solution that is both technically good and serves the political goal. In this case this is the so-called “Certificate transparency”. In a nutshell, each certificate, before being issued, is placed in a public verifiable data structure (merkle tree – used in blockchain implementations), and gets a signed certificate timestamp (SCT) as a response, which is then included as an X.509 attribute. This mitigates the risk of compromised CAs issuing certificates for websites that do not belong to the one requesting the certificate. In no time (up to 24 hours) they will be caught and distrusted, which raises the bar significantly.

Unfortunately, ETSI hasn’t included Certificate transparency in the current QWAC standard. The ESD association mentioned above says in another document that “The Browsers can easily bring any additional rules they want to impose on QTSPs such as Certificate Transparency to ETSI and other international standards bodies to be adopted through an open process of consensus by the internet community”, which I think is the wrong approach – Certificate transparency is an IETF RFC and is a de-facto standard. What will surely help is moving it (they are actually 2 RFCs) out of the “Experimental” status in IETF, but we can’t require every standard to be mirrored by ETSI.

I don’t know why CT has not been referenced by ETSI so far. It’s true that certificate log servers are based mostly in the US (and one in China), but nothing stops an organization from running a CT log. It “just” takes some infrastructure and bandwidth to support the load, but I think it’s a price EU QTSPs can pay, e.g. by sharing the costs for a couple of CT logs.

As a sidenote, I think it’s worth noting that the EU can also do more towards the adoption of DANE – a standard, which gets rid of CAs, as the public key is stored in a DNS record. It relies on DNSSEC, which doesn’t have huge adoption yet, and both are trickier to implement than it sounds, but if we want to be independent from browser decisions on which CA to trust, we can remove CAs from the trust equation. I’m fully aware it’s far from simple, and we’ll have to support PKI/CAs for a long, long time, but it’s a valid policy direction – mandate DNSSEC and DANE support.

Conclusion

Certificate transparency requirements can be added now to the eIDAS Annex IV. We are too late in the legislation process for that to be done smoothly, but I’d appreciate if the legislative bodies tried it. It would be just one additional point – “(k) details about the inclusion of the certificate in a public ledger” (note that the same eIDAS 2.0 regulates ledgers, and a CT log is a ledger, so that can be leveraged).

If not in Annex IV, then I’d strongly suggest including CT in the next version of the relevant ETSI standard. And I think it would be good if the EU Commission or ETSI to do a gap analysis of current root programs and the current ETSI standards to see if something important is missing.

Furthermore, the European Commission should initiate a series of objective studies on the effectiveness of extended validation certificates. The studies that I’ve read are not in favour of the EV/QWAC approach, but if we are to argue EV/QWACs are worth it, we need better justification.

A compromise is possible, that would make browsers confident that there will be no rogue CAs while at the same time giving Europe a say over trust in the web.

The post eIDAS 2.0, QWACs And The Security Of The Web appeared first on Bozho's tech blog.

MERDA – A Framework For Countering Disinformation

от Божидар Божанов

лиценз CC BY

Yesterday, on an conference about disinformation, I jokingly coined the acronym MERDA (Monitor, Educate, React, Disrupt, Adapt) for countering disinformation. Now I’ll put the pretentious label “framework” and describe what I mean by that. While this may not seem a very technical topic, fit for a techblog, in fact it has a lot of technical aspects, as disinformation today is spread through technical means (social networks, anonymous websites, messengers). And therefore especially the “Disrupt” part is quite technical.

Monitor – in order to tackle disinformation narratives, we need to monitor them. This includes media monitoring tools (including social media) and building reports on rising narratives that may potentially be disinformation campaigns. These tools include a lot of scraping online content, and consuming APIs where such exist and are accessible. Notably, Facebook removed much of their API access to content, which makes it harder to monitor for trends. It has to be noted that this doesn’t mean monitoring individuals – it’s just about trends, keywords, phrases – sometimes known, sometimes unknown (e.g. the tool can look for very popular tweets, extract the key phrases from it, and then search for that). Governments can list their “named entities” and keep track of narratives/keywords/phrases relating to these named entities (ministers, prime minister, ministries, parties, etc.)

Educate – media literacy, and social media literacy, is a skill. Knowing that “Your page will be disabled if you don’t click here” is a skill. Being able to recognize logical fallacies and propaganda techniques is also a skill and it needs to be taught. Ultimately, the best defense against disinformation is a well informed and prepared public.

React – public institutions need to know how and when to react to certain narratives. It helps if they know them (through monitoring), but they need the so called “strategic communications” in order to respond adequately to disinformation about current events, debunking, pre-bunking and giving the official angle (note that I’m not saying the official angle is always right – it sometimes isn’t, that’s why it has to be supported by credible evidence).

Disrupt – this is the hard part – how to disrupt disinformation campaigns. How to identify and disable troll farms, which engage in coordinated inauthentic behavior – sharing, liking, commenting, cross-posting in groups – creating an artificial buzz around a topic. Facebook is, I think, quite bad at that – this is why I have proposed a local legislation that requires following certain guidelines for identifying troll farms (groups of fake accounts). Then we need a mechanism to take them down, which takes into account freedom of speech – i.e. the possibility that someone is not, in fact, a troll, but merely a misled observer. Fortunately, the digital services act provides for out-of-court appeals for moderator decisions.

The “disrupt” part is not just about troll farms – it’s about fake websites as well. Tracking linked websites, identifying the flow of narratives through these websites, trying to find the ultimate owners, is a hard and quite technical task. We know that there are thousands such anonymous websites that repost, in various languages, disinformation narratives – but taking down a website requires good legal reasons. “I don’t like their articles” is not a good reason.

The “disrupt” part also needs to tackle ad networks – some obscure ad networks are the way disinformation websites get financial support. They usually advertise not-so-legal products. Stopping the inflow of money is one way to reduce disinformation.

Adapt – threat actors in the disinformation space (usually nation-states like Russia) are dynamic and they change their tactics, techniques and procedures (TTPs). Institutions that are trying to reduce the harm of disinformation also need to be adaptable, to constantly look for new ways of getting the false or misleading information through.

Tackling disinformation is walking on thin ice. A wrong step may be seen as curbing free speech. But if we analyze patterns and techniques, rather than content itself, then we are on mostly on the safe side – it doesn’t matter what the article says, if it’s shared by 100 fake accounts and the website is supported by ads of illegal drugs that use deep fakes of famous physicians.

And it’s a complicated technical task – I’ve seen companies claiming they identify troll farms, rings of fake news website, etc. But I haven’t seen any tool that’s good enough. And MERDA … is the situation we are in – active, coordinated exploitation of misleading and incorrect information for political and geopolitical purposes.

The post MERDA – A Framework For Countering Disinformation appeared first on Bozho's tech blog.

Anticorruption Principles For Public Sector Information Systems

от Божидар Божанов

лиценз CC BY

As a public official, I’ve put a lot of though on how to make the current and upcoming public government information systems prone to corruption. And I can list several main principles, some of them very technical, which, if followed, would guarantee that the information systems themselves achieve two properties:

they prevent paper-based corruption
they do not generate additional risk for corruption

So here are the principles that each information system should follow:

Auditability – the software must allow for proper external audits. This means having the up-to-date source code available, especially for custom-built software. If it’s proprietary, it means “code available” contract clauses. This also means availability of documentation – what components it has, what integrations exist, what network and firewall rules are needed. If you can’t audit a system, it surely generates corruption
Traceability – every meaningful action, performed by users of the system, should be logged. This means a full audit log not just for the application, but also for the underlying database as well as servers. If “delete entry” is logged at the application, but DELETE FROM is not logged by the database, we are simply shifting the corruption motives to more technically skilled people. I’ve seen examples of turned-off DB audit logs, and systems that (deliberately?) miss to log some important user actions. Corruption is thus built in the system or the configuration of its parts.
Tamper-evidence – audit logs and in some cases core data should be tamper-evident. That means that any modification to past data should be detectable upon inspection (included scheduled inspections). One of the strong aspects of blockchain is the markle trees and hash chains it uses to guarantee tamper-evidence. A similar cryptographic approach must be applied to public systems, otherwise we are shifting the corruption incentive to those who can alter the audit log.
Legally sound use of cryptography – merkle trees are not legally defined, but other cryptographic tools are – trusted timestamps and digital signatures. Any document (or data) that carries legal meaning should be timestamped with the so called “(qualified) timestamp” according to the eIDAS EU regulation. Every document that needs a signature should be signed by an electronic signature (which is the legal name for the cryptographic term “digital signatures”). Private keys should always be stored on HSMs or smartcards to make sure they cannot leak. This prevents corruption as you can’t really spoof singatures or backdate documents. Backdating in particular is a common theme in corruption schemes, and a trusted cryptographic timestamp prevents that entirely.
Identity and access management – traceability is great if you are sure you are “tracing” the right people. If identity and access management isn’t properly handled, impersonation, bruteforce or leaked credentials can make it easier for malicious internal (or external) actors to do improper stuff and frame someone else. It’s highly recommended to use 2FA, and possibly hardware tokens. For sysadmins it’s a must to use a privileged access management system (PAM).
Data protection (encryption, backup management) – government data is sometimes sensitive – population registers, healthcare databases, taxes and customs databases, etc. They should not leak (captain obvious). Data leak prevention is a whole field, but I’d pinpoint two obvious aspects. The first is live data encryption – if you encrypt data granularly, and require decryption on the fly, you can centralize data access and therefore log every access. Otherwise, if the data in the database is in plaintext, there’s always a way to get it out somehow (Database activity monitoring (DAM) tools may help, of course). The second aspect is backup management – even if your production data is properly protected, encrypted, DAM’ed, your backup may leak. Therefore backup encryption is also important, and the decryption keys should be kept securely (ideally, wrapped by an HSM). How is data protection related to corruption? Well, these databases are sold on the black market, “privileged access” to sensitive data may be sold to certain people.
Transparency – every piece of data that should not be protected, should be public. The more open data and public documents there are, the less likely it is for someone to try to manipulate data. If the published data says something, you can’t go and remove it, hoping nobody would see it.
Randomness – some systems rely on randomness for a core feature – assigning cases. This is true for courts and for agencies who do inspections – you should randomly select a judge, and randomly assign someone to do an inspection. If you don’t have proper, audited, secure randomness, this can be abused (and it has been abused many times), e.g. to get the “right” judge in a sensitive case. We are now proposing a proper random case assignment system for the judiciary in my country. It should be made sure that /dev/random is not modified, and a distributed, cryptographically-backed random-generation system can be deployed. It sounds like too much complexity just for a RNG, but sometimes it’s very important to rely on non-controlled randomness (even if it’s pseudorandomness)
Data validation – data should be subject to the maximum validation on entry. Any anomalies should be blocked from even getting into the database. Because the option for creating confusion helps corruption. For example there’s the so called “corruption cyrillic” – in countries that use the cyryllic alphabet, malicious users enter identically-looking latin charcter to hide themselves from searches and reports. Another example – in the healthcare system, reimbursement requests used to be validated post-factum. This creates incentives for corruption, for “under the table” correction of “technical mistakes” and ultimately, schemes for draining funds. If input data is validated not just a simple form inputs, but with a set of business rules, it’s less likely for deliberately incorrect data to be entered and processes
Automated risk analysis – after data is entered (by civil servants, by external parties, by citizens), in some cases risk analysis should be done. For example, we are now proposing online registration of cars. However, some cars are much more likely to be stolen than others (based on price, ease of unlocking, currently operating criminals skillset, etc.). So the registration system should take into account all known factors and require the car to be presented at the traffic police for further inspection. Similarly for healthcare – some risk analysis on anomalous events (e.g. high-price medicines sold in unlikely succession) should be flagged automatically and inspected. That risk analysis should be based on carefully crafted methodologies, put into the system with something like a rules engine (rather than hardcoded, which I’ve also seen).

Throughout the years others and myself have managed to put some of those in laws and bylaws in Bulgaria, but there hasn’t been a systematic approach to ensuring that they are all followed, and followed properly. Which is the hard part, of course. Many people know the theory, it’s just not that easy to put in in practice in a complex environment. But these principles (and probably others that I miss) need to be the rule, rather than the exception in public sector information systems if we want to reduce corruption risks.

The post Anticorruption Principles For Public Sector Information Systems appeared first on Bozho's tech blog.

Methodology for Return on Security Investment

от Божидар Божанов

лиценз CC BY

Measuring return-on-investement for security (information security/cybersecurity) has always been hard. This is a problem for both cybersecurity vendors and service providers as well as for CISOs, as they find it hard to convince the budget stakeholders why they need another pile of money for tool X.

Return on Security Investment (ROSI) has been discussed, including academically, for a while. But we haven’t yet found a sound methodology for it. I’m not proposing one either, but I wanted to mark some points for such a methodology that I think are important. Otherwise, decisions are often taken by “auditor said we need X” or “regulation says we need Y”. Which are decent reasons to buy something, but it makes security look like a black hole cost center. It’s certainly no profit center, but the more tangibility we add, the more likely investments are going to work.

I think the leading metric is “likelihood of critical incident”. Businesses are (rightly) concerned with this. They don’t care about the number of reconnaissance attempts, false positives ratios, MTTRs and other technical things. This likelihood, if properly calculated, can lead to a sum of money lost due to the incident (due to lack of availability, data loss, reputational cost, administrative fines, etc.). The problem is we can’t get company X and say “you are 20% likely to get hit because that’s the number for SMEs”. It’s likely that a number from a vendor presentation won’t ring true. So I think the following should be factored in the methodology:

Likelihood of incident per type – ransomware, DDoS, data breach, insider data manipulation, are all differently likely.
Likelihood of incident per industry – industries vary greatly in terms of hacker incentive. Apart from generic ransomware, other attacks are more likely to be targeted at the financial industry, for example, than the forestry industry. That’s why EU directives NIS and NIS2 prioritize some industries as more critical
Likelihood of incident per organization size or revenue – not all SMEs and not all large enterprises are the same – the number of employees and their qualification may mean increased or decreased risk; company revenue may make it stand out ontop of the target list (or at the bottom)
Likelihood of incident per team size and skill – if you have one IT guy doing printers and security, it’s more likely to get hit by a critical incident than if you have a SOC team. Sounds obvious, but it’s a spectrum, and probably one with diminishing returns, especially for SMEs
Likelihood of incident per available security products – if you have nothing installed, you are more likely to get hit. If you have a simple AV, you can the basic attacks out. If you have a firewall, a SIEM/XDR, SOAR, threat intel subscriptions, things are different. Having them, of course, doesn’t mean they are properly deployed, but the types of tools matter in the ballpark calculations

How to get that data – I’m sure someone collects it. If nobody does, governments should. Such metrics are important for security decisions and therefore for the overall security of the ecosystem.

The post Methodology for Return on Security Investment appeared first on Bozho's tech blog.

Why I’m Not So Alarmed About AI And Jobs

от Божидар Божанов

лиценз CC BY

With the advances in large language models (e.g. ChatGPT), referred to as AI, concerns are rising about a sweeping loss of jobs because of the new tools. Some claim jobs will be completely replaced, others claim that jobs will be cut because of a significant increase in efficiency. Labour parties and unions are organizing conferences about the future of jobs, universal basic income, etc.

These concerns are valid and these debates should be held. I’m addressing this post to the more extreme alarmists and not trying to diminish the rapid changes that these technological advances are bringing. We have to think about regulations, ethical AI and safeguards. And the recent advances are pushing us in that direction, which is good.

But in technology we often live the phrase “when you have a hammer, everything looks like a nail”. The blockchain revolution didn’t happen, and so I think we are a bit more eager than warranted about the recent advances in AI. Let me address three aspects:

First – automation. The claim is, AI will swiftly automate a lot of jobs and many people with fall out of the labour market. The reality is that GPT/LLMs should be integrated in existing business processes. If regular automation hasn’t already killed those jobs, AI won’t do it so quickly. If an organization doesn’t use automation already for boilerplate tasks, it won’t overnight automate them with AI. Let me remind you that RPA (Robotic process automation) solutions have been advertised as AI. They really “kill” jobs in the enterprise. They’ve been around for nearly two decades and we haven’t heard a large alarmed choir about RPA. I’m aware there is a significant difference in LLMs and RPA, but the idea that a piece of technology will swiftly lead to staff reduction across industries is not something I agree with.

Second – efficiency. Especially in software development, where products like Copilot are already production-ready, it seems that with the increase of efficiency, there may be staff reduction. But if a piece of software used to be built for 6 months before AI, it will be built for, say, 3 months with AI. Note that code writing speed is not the only aspect of software development – other overhead and blockers will continue to exist – requirement clarifications, customer feedback, architecture decisions, operational and scalability issues, etc., so increase in efficiency is unlikely to be orders of magnitude. AT the same time, there is a shortage of software developers. With the advances of AI, there will be less of a shortage, meaning more software can be built within the same timeframe.

For outsourcing this means that the price per hour or per finished product may increase because of AI (speed is also a factor in pricing). A company will be able to service more customers for a given time. And there’s certainly a lot of demand for digital transformation. For product companies this increase in efficiency will mean faster time-to-market for the product and new features. Which will make product companies more competitive. In both cases, AI is unlikely to kills jobs in the near future.

Sure, ChatGPT can write a website. You can create a free website with site-builders even today. And this hasn’t killed web developers. It just makes the easiest websites cheaper. By the way, building software once and maintaining it are completely different things. Even if ChatGPT can build a website, maintenance is going to be tough through prompts.

At the same time, AI will put more intellectual pressure on junior developers, who are typically given the boilerplate work, which is going to be more automatable. But on the other hand AI will improve the training process of those junior developers. Companies may have to put more effort in training developers, and career paths may have to be adjusted, but it’s unlikely that the demand for software developers will drop.

Third, there is a claim that generative AI will kill jobs in the creative professions. Ever since I wrote an algorthmic music generator, I’ve been saying that it will not. Sure, composers of elevator music will be eventually gone. But poets, for example, won’t. ChatGPT is rather bad at poetry. It can’t actually write proper poetry. It seems to know just the AABB rhyme scheme, it ignores instructions on meter (“use dactylic tetrameter” doesn’t seem to mean anything to it). With image and video generation, the problem with unrealistic hands and fingers (and similar ones) doesn’t seem to be going away with larger models (even though the latest version of Midjourney is neatly going around it). It will certainly require post-editing. Will it make certain industries more efficient? Yes, which will allow them to produce more content for a given time. Will there be enough demand? I can’t say. The market will decide.

LLMs and AI will be change things. It will improve efficiency. It will disrupt some industries. And we have to debate this. But we still have time.

The post Why I’m Not So Alarmed About AI And Jobs appeared first on Bozho's tech blog.