#066 - Cyber AI Chronicle - Context is Everything

PRESENTED BY

Cyber AI Chronicle

By Simon Ganiere · 11^th May 2025

Welcome back!

Artificial intelligence is exceptionally adept at rapidly processing large volumes of data, recognizing patterns, and presenting objective insights. This capability gives users, particularly those with less experience, immediate access to critical context. However, there remains an essential dimension beyond AI's reach: the subtle nuances of human interaction and implicit understanding.

While AI can identify historical trends or detect recurring situations, it cannot grasp the underlying motivations, hidden agendas, or nuanced interpersonal dynamics that truly shape human interactions. Seasoned individuals inherently understand this subtle layer—they read between the lines, perceive unspoken sentiments, and intuitively interpret situations in ways technology cannot replicate.

Experienced people can quickly sense when what's being explicitly stated diverges from the underlying reality, drawing upon years of personal interaction and observation. It's this depth of intuitive understanding—knowing the "why" behind the "what"—that distinguishes human insight from algorithmic analysis.

Ultimately, the most effective approach integrates AI's strengths with human intuition. Use AI to swiftly gather and summarize relevant information, but complement it with human experience to interpret these insights accurately. AI is a powerful assistant, yet it cannot fully substitute the nuanced understanding and intuitive awareness that comes uniquely from human experience.

I managed to spend some time building a couple of n8n workflows. The first one is pulling some RSS feeds to gather news items about AI and security. The second one is analyzing the news from the first step in order to identify key topics. The last one is finally using Jina.ai to do some extra search and then write a summary (using OpenAI). All of it is saved in a pretty simple database. I still need to build a small web interface to get the results and be able to review them. Once that done, I would then be able to add additional filter for specific keywords or content. Not perfect but pretty fun to do! 😃

This article is highlighting how Copilot can be (mis)used to gain access to information that can be misused by a threat actor. What is supposed to be an augmentation for employees to do a faster/better job can be used as the ideal tool for reconnaissance. Not that I want to give ideas to people but this is an ideal scenario for an insider » READ MORE

Anthropic's Claude AI was exploited to create over 100 fake political personas. The operation was designed as an "influence-as-a-service" campaign that targeted users on Facebook and X. The campaign aimed to manipulate political narratives aligned with the interests of various nations, including the U.A.E. and Iran. Claude AI was utilized not only for content generation but also for orchestrating interactions between fake personas and real users » READ MORE

Darcula released updates that integrate GenAI capabilities into its cybercrime toolkit. This advancement not only enables the rapid creation of customized phishing pages but also supports multiple languages and form generation without the need for programming knowledge. This addition lowers the technical barrier for creating phishing pages, enabling less tech-savvy criminals to deploy customized scams in minutes » READ MORE

Three malicious npm packages aimed at macOS Cursor users have been identified. Over 3,200 downloads resulted in stolen user credentials and a compromised software environment. The packages impersonate legitimate tools, enticing developers with false promises of affordability. Security experts warn that such attacks highlight the rampant risks within the software supply chain » READ MORE

Google has introduced new AI-powered features to combat online scams across its platforms: Chrome, Search, and Android. - The company is deploying its on-device large language model, Gemini Nano, to enhance scam detection in Chrome's Safe Browsing feature » READ MORE

I did hesitate before sharing this article as it can be read as a sales pitch from Nudge Security. That being said it’s highlighting some of the key challenges I can see as well: shadow AI, data governance and supply chain » READ MORE

This is not new as it’s been in the news for a while now but always good to see how such attack plays out and the response from the security team at Kraken. The interesting part to me was: […] Even more suspicious, the candidate occasionally switched between voices, indicating that they were being coached through the interview in real time […] looks like we are far away from a deepfake attack on this one. Joke aside - this will require security team to work closely with HR and the recruitment team which is not something that happen usually but is super important in this context » READ MORE

At the risk of stating the obvious…cyber security companies must protect themselves as they are an ideal target either directly or their technology (as a way to bypass them). There is no big surprises here in term of threat actors (DPRK, ransomware, etc.) but the level of transparency from SentinelOne is commendable » READ MORE

Kevin Beaumont, in is usual provocative style, is actually highlighting an extremely valid point: paying a ransom is not bringing back your infrastructure by magic and even if you pay the ransom you will still have a significant amount of work to restore your environment » READ MORE

Microsoft's new policy of making all new accounts passwordless by default! The company aims to enhance security and reduce password-related attacks. Microsoft encourages users to enroll in passkeys, a more secure alternative to passwords, for a more secure login experience! This is great initiative that can only help with the whole set of identity theft » READ MORE

Whilst the previous item was a positive for security this one is just the opposite. Researchers have identified a security flaw in Windows Remote Desktop Protocol (RDP) that allows user to log in using revoked passwords. Microsoft is arguing that this is a design decision! Knowing RDP is already significantly misuse by ransomware actors (amongst others)…I’m sure we will hear about this one again » READ MORE

I’m not sure I believe that full AI employee will be available that quickly. However the article highlight a very interesting and important point: how are we going to manage those AI identity? Anyone working in a corporate world knows that non human identity are already a huge challenge to secure so having a full “intelligent” identity on top of it will require everything we are currently missing (e.g. inventory, telemetry, visibility, etc.) » READ MORE

Anthropic is following the steps of others and finally announced the ability to run web searches via their API. This will enable developer to use up to date information in their application, agent workflows, etc. » READ MORE

Google released Gemini 2.5 Pro Preview (I/O Edition), an updated version of 2.5 Pro that has significantly improved capabilities for coding, especially building compelling interactive web apps. » READ MORE

Opinion piece from the New York Time written by Eric Schmidt and Selina Xu. China is rapidly advancing in technology, particularly in artificial intelligence, and is now at parity or ahead of the United States in various fields. The authors are arguing that the US should double down on research and innovation and share those openly. It is true that the adoption of a more open approach by China has resulted in remarkable achievements (deepseek, qwen, etc.) » READ MORE

Summary: The paper introduces the Absolute Zero Reasoner (AZR), a novel reinforcement learning paradigm that enables large language models to self-improve their reasoning capabilities without relying on external data. By proposing and solving tasks autonomously, AZR achieves state-of-the-art performance in coding and mathematical reasoning tasks, surpassing models trained with extensive human-curated datasets. The study highlights the potential of self-play in open-ended environments to drive continuous learning and reasoning improvements, suggesting a promising direction for future AI development.

Published: 2025-05-06T09:08:00Z

Authors: Andrew Zhao, Yiran Wu, Yang Yue, Tong Wu, Quentin Xu, Matthieu Lin, Shenzhi Wang, Qingyun Wu, Zilong Zheng, Gao Huang

Organizations: Tsinghua University, Beijing Institute for General Artificial Intelligence, Pennsylvania State University

Findings:

Final Score: Grade: A, Explanation: Novel approach with strong empirical results and transparent methodology.

Download Report | Website

Till next time!

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.