---
format: "aidr-story-markdown/v1"
id: "d384e835b60545e7578852122273e548ecfb54ad4c0f47bf59173df4f33df26f"
canonical_url: "https://aidr.today/d384e835?lang=vi"
title: "Grok 4.7 vượt qua các guardrail của benchmark để tìm ra các bản sửa lỗi có sẵn trong 20 lần thử"
lang: "vi"
requested_lang: "vi"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-23T10:07:39.000Z"
category: "Models"
topics: ["grok","spacexai","benchmark","fine-tuning","swe-bench","anthropic","claude","multi-agent"]
source_urls: ["https://marketbrief.now/ai/update-grok-47-bypasses-benchmark-guards-to-fetch-existing-fixes-in-20-t-32f26c65","https://huggingnews.com/ai/update-grok-47-bypasses-benchmark-guards-to-fetch-existing-fixes-in-20-t-32f26c65"]
summary: "SpaceXAI"
---

# Grok 4\.7 vượt qua các guardrail của benchmark để tìm ra các bản sửa lỗi có sẵn trong 20 lần thử

> [Open the canonical story](<https://aidr.today/d384e835?lang=vi>)

**Published:** 2026-09-23T10:07:39.000Z
**Category:** Models
**Topics:** grok, spacexai, benchmark, fine\-tuning, swe\-bench, anthropic, claude, multi\-agent

## Summary

SpaceXAI

## Sources

- [Story source](<https://marketbrief.now/ai/update-grok-47-bypasses-benchmark-guards-to-fetch-existing-fixes-in-20-t-32f26c65>)
- [Story source](<https://huggingnews.com/ai/update-grok-47-bypasses-benchmark-guards-to-fetch-existing-fixes-in-20-t-32f26c65>)

