1. The US Congressional Budget Office confirmed unauthorized ac…
The US Congressional Budget Office confirmed unauthorized access to its networks in a suspected foreign cyberattack potentially exposing emails with congressional offices.
Category: Newsletter
2. A significant data leak from Chinese cybersecurity firm Know…
A significant data leak from Chinese cybersecurity firm Knownsec exposed cyber weapon specifications, proprietary hacking tool source code, government collaboration details, and a list of 80 compromised overseas surveillance targets.
Category: Newsletter
3. Meta’s chief AI scientist Yann LeCun is preparing to leave t…
Meta’s chief AI scientist Yann LeCun is preparing to leave to start a venture focused on world models, amid Meta’s AI reorganization including Meta Superintelligence Labs, and internal tensions over AI strategy shifts.
Category: Newsletter
4. Bytedance launched the Doubao-Seed-Code AI coding agent pric…
Bytedance launched the Doubao-Seed-Code AI coding agent priced at 40 yuan/month, with a launch discount to 9.9 yuan (US$1.30). The model sets records on SWE-Bench Verified tests, supports popular dev tools, and processes up to 256,000 words per query for complex codebases.
Category: Newsletter
5. This detailed guide explains setting up pre-training infrast…
This detailed guide explains setting up pre-training infrastructure and training a 1B parameter Llama 3-style model on 8xH100 GPUs, providing a working implementation for further abstraction, though it is not state-of-the-art.
Category: Newsletter
6. Researchers evaluated nine AI models in a simulated workplac…
Researchers evaluated nine AI models in a simulated workplace with 150 tasks, finding most had limited coherence and lacked common sense, suggesting fundamental capabilities must be acquired before reliable real-world reasoning is achievable.
Category: Newsletter
7. Sudoku variants test AI creativity with rule modifications a…
Sudoku variants test AI creativity with rule modifications absent from training data. GPT-5 is the first LLM to solve a 9×9 variant coordinating four constraint types simultaneously but succeeded only 20% of the time.
Category: Newsletter
8. Current LLMs lack spatial intelligence, a key human cognitive function
Current LLMs lack spatial intelligence, a key human cognitive function. Emerging “world models” generate physically consistent simulated environments using multimodal inputs, potentially advancing robotics, video modeling, and drug discovery significantly.
Category: Newsletter
9. The agentic coding field has seen rapid development with par…
The agentic coding field has seen rapid development with paradigm shifts over ten months. Next year promises polished systems integrating new-age models focused on advanced context management to translate intent into functioning code efficiently.
Category: Newsletter
10. This comprehensive whitepaper by Google researchers introduc…
This comprehensive whitepaper by Google researchers introduces AI agents, which are autonomous applications capable of planning and task execution, combining reasoning capabilities with action-oriented functions.
Category: Newsletter
11. Anthropic released a searchable library featuring over 45 us…
Anthropic released a searchable library featuring over 45 use cases for its Claude AI, covering professional, marketing, educational, research, and personal contexts, with detailed prompts and step-by-step instructions.
Category: Newsletter
12. The Quantum Route Redirect phishing-as-a-service platform us…
The Quantum Route Redirect phishing-as-a-service platform uses around 1,000 pre-configured domains and automated bot filtering to steal Microsoft 365 credentials in 90 countries, mainly the US, by redirecting humans to credential-harvesting pages while evading security tools.
Category: Newsletter
13. GlobalLogic suffered a data breach exposing personal and fin…
GlobalLogic suffered a data breach exposing personal and financial data of over 10,000 employees. The Clop ransomware group exploited Oracle E-Business Suite vulnerabilities, a known tactic involving data theft and payment coercion. Emergency patches have been released.
Category: Newsletter
14. Researchers revealed a zero-day vulnerability in Gladinet’s …
Researchers revealed a zero-day vulnerability in Gladinet’s Triofox platform, permitting unauthenticated attackers to impersonate ‘localhost’ via HTTP host header manipulation and replace anti-virus scanner paths with malicious scripts, allowing arbitrary code execution.
Category: Newsletter
15. The GlassWorm malware resurfaced post-eradication, infecting…
The GlassWorm malware resurfaced post-eradication, infecting OpenVSX VS Code extensions and GitHub repositories through complex attacks using invisible Unicode and blockchain-based command infrastructure, stealing credentials and pushing malicious AI-generated commits.
Category: Newsletter