Kimi K3 placed second in Firmulate’s company management test, ahead of three frontier models. The results put follow-through, file reading and trust in focus.
The AI Model League Is More Open Than the Chat Demos Suggest
Browsing Category
AI for Cybersecurity
154 posts
Claude Code Reads AGENTS.md Only When Telemetry Is On [Fixed]
![Claude Code Reads AGENTS.md Only When Telemetry Is On [Fixed]](https://smartcr.org/wp-content/uploads/2026/09/claude-code-reads-agents-md-only-when-telemetry-is-on-fixed-featured.jpg)
Recent updates show Claude Code accesses AGENTS.md exclusively when telemetry is active, raising questions about data handling and privacy.
The AI Security Experiment: Using Anthropic’s Claude To Access OpenAI
Security researchers reportedly used Anthropic’s Claude AI to breach an OpenAI system, raising concerns over AI-enabled cyberattacks and industry safety measures.
ChatGPT Now Knows What You Do On Other Websites Via Ad Collector
Recent trend signals suggest ChatGPT may now be aware of user activity on other websites through ad tracking, raising privacy concerns amid rising coverage interest.
OpenAI ‘Ethically Hacked’ With Help Of Anthropic’s Claude Chatbot
OpenAI reportedly engaged in an ‘ethical hacking’ exercise with Anthropic’s Claude chatbot, raising questions about AI safety testing and inter-company collaboration.
Why the Worst AI Manager in the League Still Scores 26: Inside an Honest Benchmark
One AI benchmark gives a do-nothing manager 26 points, caps grades after a single trust breach, and trusts no perfect score. Here’s the business logic.
Introducing Astra For Law
Astra for Law, a new AI-powered legal platform, has been announced, aiming to streamline legal research and case analysis. Details are still emerging.
AI Misuse In September 2026: How Anthropic Is Tackling The Challenge
Anthropic’s September 2026 report details efforts to detect and counter AI misuse, highlighting ongoing safety measures amid evolving threats.
Three Key AI Warnings That Were Almost Overlooked
A detailed analysis of three overlooked AI threats, based on recent investigations into OpenAI’s security incidents and emerging agent capabilities.
Diligence Without Delivery: What Opus 4.8 Teaches Us About AI at Work
Opus 4.8 learned 80-plus rules and produced the deepest analysis, yet finished last—showing why disciplined execution matters more than volume.
