Unverified50% confidenceEventExact time
When researchers instructed Claude Code, Gemini, and ChatGPT to autonomously search for a certain capability, they all directly found the malicious repositories.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/25/2026
First Seen
Valid until: 12/24/2026
Sources
Related Entities
Related Claims
UnverifiedThe new generation of programming agents represented by Cursor, Devin, Claude Code, and Gemini CLI can not only generate code but also autonomously invoke terminal commands, read and write files, run tests, and iteratively fix errors based on error messages.64% similarUnverifiedA technical blog post reported that Claude Code appears to read the AGENTS.md file in the project root directory only when the telemetry feature is enabled.64% similarUnverifiedClaude Code's sub-agent design isolates exploration tasks from modification tasks, allowing the main agent to dispatch multiple independent Claude instances in parallel for exploration.63% similarUnverifiedUsing non-Anthropic models in Claude Code is not easy, often requiring hacks and being prone to bugs.62% similarUnverifiedIn the experiment, Claude mistook a planned tool that existed only in the specification document for an actually installed part of the technology stack.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/948684API
curl https://kongchang.com/api/v1/knowledge/claims/948684MCP
get_claim(id=948684)