#aisi

UNWIRE.HK @unwirehk_mirror@mastodon.hongkongers.net

· 3mo ago

AI 測試首度攻擊真人 Anthropic 模型偽造身份施壓開發者英國政府人工智能安全研究所（AISI）於 8 月 4 日發布報告，指近期一項網絡安全評估測試中，部分 AI 智 […] #人工智能 #AISI #Anthropic #GPT-5.6 Sol

unwire.hk

AI 測試首度攻擊真人 Anthropic 模型偽造身份施壓開發者

在這 19 宗事件中，17 宗涉及 Anthropic Mythos 5 模型，2 宗涉及 OpenAI GPT-5.6-Sol 模型。測試期間，兩間公司均刻意關閉各自分類過濾器（即實時攔截危險請求安全機制）並開放互聯網連接，目的是了解模型原始能力，並非因為系統遭突破。

Xavier Ashe

@Xavier@infosec.exchange

· 3mo ago

First OpenAI, the Anthropic, now #AISI let's an AI hack things, then releases an incident report/press release.

Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work

aisi.gov.uk

Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work

During a routine cyber evaluation, AISI identified an incident in which AI agents took sustained, unsanctioned action directed at real people and organisations. We are disclosing what we found, what it means, and the actions now underway.

You've seen all posts