PRESENTED BY
Cyber AI Chronicle
By Simon Ganiere · 26th May 2024
Welcome back!
Project Overwatch is a cutting-edge newsletter at the intersection of cybersecurity, AI, technology, and resilience, designed to navigate the complexities of our rapidly evolving digital landscape. It delivers insightful analysis and actionable intelligence, empowering you to stay ahead in a world where staying informed is not just an option, but a necessity.
Table of Contents
What I learned this week
TL;DR
Microsoft is in the news..again…and the security teams are raising an eyebrow (if not both). After having disclosed that security is a top priority…they release a function named Recall…which basically takes a screenshot every five(!) seconds and then leverage a local AI model to be able to search that content. Don’t get me wrong Microsoft is not the only one. Slack has been facing a significant backslash as well. Their policy includes the ability to ingest customers’ data to train their model. Yes, that means your message, content and files.
I wanted to deep dive on those topics, but more particularly on the Microsoft Recall announcement. » READ MOREGoogle seems to be doing Google things…they feel the pressure from the market in terms of AI and like in the past with social media they are trying to add AI feature everywhere. The latest being that AI Overview in the search results…and obviously not working brilliantly as illustrated by the recommendation to add some glue to make cheese stick on pizza…as I mentioned previously not everything has to be AI driven 😉
On the security side, another week and another bunch Microsoft and Google Chrome of vulnerabilities being actively exploited that you have hopefully already patched (if you haven’t please go apply the patches and then come back to finish reading this newsletter). Still on the vulnerability management topic, some really worrying news on the status of the NVD. I did mention this in the past already but the latest stats are really not looking great. A whooping 93% of the new vulnerabilities have NOT been analysed but the NVD since February. » READ MORE
On the back of last week tutorial, I have now added a section Research Paper. Obviously leveraging the previous work. Will check if I keep this every week or only when there is interesting research paper.
In a lighter but sad note, ICQ is shutting down…not making me feel any younger but definitely nostalgic 😔
AI Technology Can Increase The Attack Surface
As mentioned in the introduction, Microsoft has been in the news this week (again) for the Recall feature. I wanted to zoom into this from a security perspective and how this new feature is most probably increasing the attack surface.
What is Recall?
As per Microsoft website, […] Search across time to find the content you need. Then, re-engage with it. With Recall, you have an explorable timeline of your PC’s past. Just describe how you remember it and Recall will retrieve the moment you saw it. Any photo, link, or message can be a fresh point to continue from. As you use your PC, Recall takes snapshots of your screen. Snapshots are taken every five seconds while content on the screen is different from the previous snapshot. Your snapshots are then locally stored and locally analyzed on your PC. Recall’s analysis allows you to search for content, including both images and text, using natural language. […]
Also note you need to have a Copilot+ PC.
Sounds all nice and (somewhat) interesting. From this everything is done locally and not uploaded to Microsoft cloud or any third party…so what’s the problem?
Privacy nightmare
Whilst the fact that Recall is not taking screenshots of private browsing and you can block some websites is another positive sign, the problem here is that by taking screenshots every 5 seconds, it's capturing more or less everything you do and harvesting even more data. You can get everything by doing this: emails, browser, files open, etc. This means that any scenario of unauthorized access will be leading to access to more or less all of your activity. Knowing also some of the track record of Microsoft, trusting them with something like this is a lot to ask. The Information Commissioner's Office in the UK (ICO) has already made enquiries to Microsoft.
How threat actors will leverage this?
I can think of a couple of scenarios where a threat actor can hit the jackpot. The obvious one is a threat actor dropping an info stealer type of malware. Usually, those malware try to steal identities and credentials, for which you can argue Recall will not capture as normally passwords should be hidden by default - that being said, I can think of a few situations with password managers where people are looking at passwords in clear text. So if that situation is not fully relevant, what's the problem?
As per Microsoft, Recall will not do content moderation. So if you have that password in clear text, or that credit card number, or that bank account number, etc., it will be captured and potentially available to a threat actor. This augments significantly the attack surface. By creating a single stop shop where everything is recorded, it's making the life of a threat actor a lot easier. Now if it wasn't bad enough, here is the kicker:
Yep, you read that right…a basic folder with screenshots and SQLite database…any sort of basic info stealer malware can take that and send it over the internet. That entire dataset can then become the perfect source for social engineering. Knowing you have a snapshot over a long time of the activity of someone, you can then customize that phishing email very accurately.
It remains a mystery to me why this data is at least not encrypted.
Also, the other scenario that is overlooked is physical access. I'm thinking about abuse situations where the abuser could then leverage access to Recall to basically search for any activities and use this as a way to continue their abuse.
Conclusion & Recommendations
I would strongly recommend thinking twice before enabling such a feature. Yes, it might be useful for some, but it creates a significant attack surface that should not be neglected. In enterprise settings, this should most probably be disabled altogether. For home users, I would strongly recommend reading the Recall page and the “Privacy and control over your Recall experience” page from Microsoft so you can make an informed decision. Until Microsoft increases its security on this functionality, I would stay far away from it.
Additionally, here are some recommendations to mitigate the risks associated with the Recall feature:
Enable full disk encryption on your device to protect the Recall data in case of physical access or theft.
Use a reputable anti-malware solution and keep it up-to-date to detect and prevent potential info-stealer malware attacks. Keep your operating system up-to-date as well.
Consider using a virtual machine or a separate user account for sensitive activities to isolate the Recall data from other activities.
Regularly review the Recall data and delete any sensitive information that is no longer needed.
Advocate for Microsoft to implement stronger security measures, such as encryption and access controls, for the Recall data.
This is a reminder that the adoption of any new technology also create risks and we should not be blinded by adopting the latest technology. This applies particularly to AI, and applying even a basic risk management approach to any new AI deployment or feature is the right thing to do.
Worth a full read
Wiz Researcher discovers critical vulnerability in AI-as-a-Service provider, Replicate.
Key Takeaway
Critical vulnerabilities in AI-as-a-service platforms can expose millions of private AI models.
Swift mitigation and transparency in handling security issues are crucial for maintaining trust.
The ease of deploying AI models comes with significant security risks from malicious actors.
Cross-tenant attacks via malicious AI models highlight the need for robust isolation measures.
Investigating cloud infrastructure security can reveal common but overlooked vulnerabilities.
The ability to manipulate data across tenants underscores the importance of secure data handling.
Security research plays a vital role in identifying and mitigating emerging threats in AI services.
The Real Danger Lurking in the NVD Backlog
Key Takeaway
The NVD backlog highlights a critical vulnerability in global cybersecurity infrastructure.
The security community must adapt quickly to mitigate risks from unanalyzed vulnerabilities.
Real-time vulnerability intelligence services are becoming essential in threat management.
The delay in analyzing vulnerabilities underscores the importance of alternative data sources.
Since February 12, 2024, over 12,720 new vulnerabilities have been added to NVD.
11,885 of these new vulnerabilities have not been analyzed or enriched by NVD.
NVD has been a primary source for software vulnerability data for over 20 years.
50.8% of Known Exploited Vulnerabilities remain unanalyzed by the NVD since February.
55.9% of Weaponized Vulnerabilities have not been analyzed by the NVD since February.
82% of CVEs with a Proof-of-Concept Exploit are unanalyzed by the NVD since February.
Proofpoint’s 2024 Voice of the CISO Report
Key Takeaway
74% of CISOs identify human error as their biggest cyber vulnerability in 2024.
Ransomware is the top concern among CISOs, reflecting an increase in sophisticated attacks.
87% of global CISOs are deploying AI-powered capabilities against human-centric cyber threats.
Economic conditions have negatively impacted 59% of organizations' ability to resource cybersecurity budgets.
66% of global CISOs express concern over personal, financial, and legal liability in their roles.
Fewer than half (46%) of global CISOs reported a material loss of sensitive information in the past year.
84% of CISOs believe cybersecurity expertise should be required at the board level.
Generative AI poses a security risk according to 54% of CISOs surveyed.
Cybersecurity budgets have been cut most severely in education (68%) and healthcare (68%).
66% of CISOs agree that expectations on them are unrealistic, an increase from previous years.
Research Paper
Smart Validator: A Framework for Automatic Identification and Classification of Cyber Threat Data
Summary: The paper presents SmartValidator, an AI-based framework for automating the validation of cyber threat alerts using machine learning techniques. It consists of three layers: data collection, model building, and alert validation. The framework dynamically constructs validation models based on Security Operation Centre (SOC) requirements and Cyber Threat Information (CTI). A Proof of Concept (PoC) system was built using eight ML algorithms and two feature engineering techniques. Evaluation results showed that 75% of the models had an F1-score above 0.8, indicating adequate performance. The dynamic construction of models required 99% fewer models than pre-building for all possible requirements, demonstrating efficiency. The framework can be adopted by various industries to accelerate and automate alert validation.
Authors: Chadni Islam, M. Ali Babar, Roland Croft, Helge Janicke
Organizations: University of Adelaide, Cyber Security Cooperative Research Centre
Findings:
SmartValidator automates alert validation using AI.
75% of models had F1-scores above 0.8.
Dynamic model construction required 99% fewer models.
Final Score: Grade: A-, Explanation: Novel approach with strong empirical support but some missing statistical details.
Download: https://arxiv.org/pdf/2203.07603v1
Some more reading
Initial Access: The 2024 Disagreement » READ
The UK to propose mandatory reporting for ransomware attacks and licensing regime for all payments » READ
Man behind deepfake Biden robocall indicted on felon charges, faces $6M fine » READ
Deepfakes rank as the second most common cybersecurity incident for US businesses » READ
Why bother creating your own encryption routine for your ransomware…when you use BitLocker » READ & READ
How major acquisitions are transforming security operations » READ
Identity vulnerabilities a concern at Microsoft, outside researcher claims » READ
Microsoft will “like” VBScript - Some companies literally run on Excel macro so better be sure you are looking at this ahead of time » READ
Wisdom of the week
Strong leaders are long on vision.
Weak leaders are short on sight.
Contact
Let me know if you have any feedback or any topics you want me to cover. You can ping me on LinkedIn or on Twitter/X. I’ll do my best to reply promptly!
Thanks! see you next week! Simon