Claude Code is implementing steganographic techniques to mark requests for privacy reasons. The inspection of the Claude Code binary revealed concerns regarding the permissions granted to coding agents, which often include extensive access to files, shells, and repositories.

Copy Fail is a logic flaw that allows for local privilege escalation on Linux systems without the need for a race window or kernel-specific offsets. A 732-byte Python script can exploit this flaw to gain root access on every Linux distribution released since 2017, functioning unmodified across multiple systems.

Anthropic Mythos's showcase vulnerabilities were tested on small, inexpensive, open-weight models, revealing similar analysis results. AI cybersecurity capability varies significantly with model size, indicating that the security moat relies on the system architecture rather than the model itself.

Posted by morsch. Score: 633 points. Comments: 600.

GLM 5.2 from Zhipu AI achieved a 39% F1 score on the IDOR detection benchmark, outperforming Claude Code, which scored 32%. Both models were evaluated on the same dataset and prompt, with GLM 5.2 costing approximately $0.17 per vulnerability found.

74% of consumers feel the internet is less human than it was a decade ago. 61% of consumers cannot identify a brand that effectively uses AI in its messaging, while enterprise teams spend an average of 16.6 hours per week on improving AI visibility.

DeepSeek V4 Pro 0813 is hosted by a single provider with OpenRouter forwarding requests directly. Average customer prices are often lower than listed due to caching and discounts, while performance metrics include throughput (tokens per second), latency (total round-trip time), and TTFT (time-to-first-token).

ChatGPT's functionality is contingent on a Cloudflare Turnstile program that analyzes user React state before allowing input. This program examines 55 properties across three layers: browser characteristics, Cloudflare network data, and user location information.

Discovery LP automates the scientific discovery process to enhance efficiency in science and engineering. The platform addresses the limitations of traditional experimental methods, which involve repetitive and manual iterations that hinder progress.

Kimi K3 is an open model that offers competitive performance with Fable 5 at a lower cost. When combined, Kimi K3 and Fable 5 achieve state-of-the-art results across approximately 1,030 agentic tasks.