Breaking
TRExpeditions started on Çerkezköy-Kapıkule High Speed ​​Railway LineBRStorm with wind gusts of 76 km/h knocks down trees and leaves residents homeless in ResendeTRThe starting 11's have been announced!ITZelensky, bombs on Zaporizhzhia are Putin's 'thank you' for lifting sanctionsITVerona, little girl hit by a car in the courtyard of her house in Casaleone: it's seriousRUThe situation in the Kiev metro after the stop was caught on videoCNLai Qing’s Delegation Day speech: “Strengthening national defense is a deterrent, not a provocation.” The Taiwan Affairs Office of the State Council jumped on the bandwagonBROur Lady of Aparecida holiday: what opens and closes in Juiz de ForaBRRussian attacks kill 12, says Ukraine hours after agreement between Trump and PutinRULantratova spoke about negotiations on the situation in AleshkiTRExpeditions started on Çerkezköy-Kapıkule High Speed ​​Railway LineBRStorm with wind gusts of 76 km/h knocks down trees and leaves residents homeless in ResendeTRThe starting 11's have been announced!ITZelensky, bombs on Zaporizhzhia are Putin's 'thank you' for lifting sanctionsITVerona, little girl hit by a car in the courtyard of her house in Casaleone: it's seriousRUThe situation in the Kiev metro after the stop was caught on videoCNLai Qing’s Delegation Day speech: “Strengthening national defense is a deterrent, not a provocation.” The Taiwan Affairs Office of the State Council jumped on the bandwagonBROur Lady of Aparecida holiday: what opens and closes in Juiz de ForaBRRussian attacks kill 12, says Ukraine hours after agreement between Trump and PutinRULantratova spoke about negotiations on the situation in Aleshki
Back中國9大AI開發商發表857個模型 僅4.7%公布安全測試結果
中國9大AI開發商發表857個模型 僅4.7%公布安全測試結果
Tech
自由时报4 hours agoTech2 min readChinaView translation

中國9大AI開發商發表857個模型 僅4.7%公布安全測試結果

研究顯示中國9大AI業者發布的857個模型中,僅4.7%公布具體安全測試結果,凸顯產業透明度問題。

Quick Look

根據研究機構SemiAnalysis報告指出,中國9大主要AI開發商在2021年至今年9月發布的857個AI模型中,僅4.7%公布具體的安全測試結果,引發外界對先進AI風險與透明度的擔憂。

AI-generated summary

Why It Matters

全球對先進人工智慧模型可能產生的風險日益擔憂,研究機構針對AI開發商的安全測試結果透明度進行評估。

Font size

《路透》報導,全球對先進人工智慧(AI)模型可能產生的風險日益擔憂之際, 有研究機構發布報告指出,中國主要AI開發商只對少數已開發的AI模型公布具體的安全測試結果,佔比僅4.7%。

總部位於加州的科技研究機構SemiAnalysis,檢視中國9大AI業者從2021年至今年9月15日期間所發表的857個AI模型,這些業者包括阿里巴巴、字節跳動、騰訊、百度、深度求索(DeepSeek)、月之暗面(MoonShot AI)、智譜(Z. AI)、稀宇科技(MiniMax)和階躍星辰(StepFun)。

報告指出,在857個模型中,僅31個公開可與具體模型對應的安全評估結果,佔比僅3.6%;另有9個模型在發表或發表前即已公布這類結果,占比1.1%。這意味著有817個模型未揭露安全相關資訊,不過這些企業仍有可能在內部進行測試。

SemiAnalysis表示,其對「資訊揭露」的定義是指,與具名模型相關連的具體結果,包括對有害輸出、越獄抵抗能力(jailbreak resistance)、毒性(toxicity)、隱私、拒絕行為或危險能力的測試,而且不將某模型已接受安全訓練或評估等一般性聲明計算在內。

絕大多數能支援AI代理(AI agents)自主進行網路攻擊的模型,不是由美國就是由中國的開發商所研發。

《路透》上週報導,中國的AI代理在測試中展現出欺騙使用者、規避限制,以及隱瞞失敗的能力,這與外界對美國先進系統所提出的擔憂不謀而合。

Open Questions

  • 中國AI企業內部實際的安全測試頻率為何?
  • 未來是否會加強公開AI安全評估結果?

Related Topics

This article was originally published by 自由时报.

Related Stories

Various robots show "unique skills" in Wuhan, attracting buyers from more than 20 countries
Tech·

Various robots show "unique skills" in Wuhan, attracting buyers from more than 20 countries

The 2026 Wuhan Artificial Intelligence and Robotics Expo opened on the 10th. More than 200 companies participated in the exhibition with products, attracting purchasing groups from more than 20 countries. A variety of innovative robots and aircraft were unveiled and a large number of overseas intended orders were obtained, demonstrating Wuhan's advantages in the artificial intelligence and robotics industry.

中国新闻网
3 min read
More on this topic人工智慧