Claude Mythos恐威脅金融安全?美財長、聯準會主席緊急開會警告風險

AWS3,55%

美國財政部長與聯準會主席緊急召集華爾街高層,針對 Anthropic 最新模型 Mythos 示警。該模型被當局定性為金融體系的系統性風險。

財金首長緊急召集華爾街高層,AI 網路威脅升格為系統性風險

根據《彭博社》報導,美國財政部長史考特・貝森特(Scott Bessent)與聯準會主席傑洛姆・鮑爾(Jerome Powell)於上週在華府財政部總部緊急召集華爾街主要銀行執行長,就 AI 公司 Anthropic 最新模型「Claude Mythos Preview」可能引發的網路安全風險提出警告。

出席會議的銀行高層包括花旗集團執行長范潔恩(Jane Fraser)、摩根士丹利執行長泰德・皮克(Ted Pick)、美國銀行執行長布萊恩・莫伊尼漢(Brian Moynihan)、富國銀行執行長查爾斯・沙夫(Charlie Scharf),以及高盛執行長大衛・所羅門(David Solomon);摩根大通執行長傑米・戴蒙(Jamie Dimon)則未能出席。上述機構均被主管機關列為具系統重要性的金融機構。

此次會議由財政部長與聯準會主席親自出席,被業界視為異常之舉。過去政府對 AI 風險的介入,多半停留在機構層級的工作小組,如今直接升格為金融當局最高首長親自示警,明確釋出當局已將 AI 網路威脅定性為金融體系穩定性的系統性風險。

Mythos 模型能力驚人,可自主發現大量零時差漏洞

根據 Anthropic 於同日公佈的技術資料,Mythos 具備在所有主流作業系統及網頁瀏覽器中識別並利用漏洞的能力。在測試階段,該模型自主找出數千個先前未知的零時差漏洞(zero-day vulnerabilities),其中包括安全導向開源作業系統 OpenBSD 中一個潛伏長達 27 年的漏洞;此外,該模型還在影音處理函式庫 FFmpeg 中發現一個即使經過自動化測試工具執行 500 萬次程序也未能偵測的漏洞。

Anthropic 研究人員強調,Mythos 的漏洞發現能力源自模型在程式撰寫、推理及自主性方面的整體進步,並非刻意訓練的結果。

該公司在聲明中指出,「讓模型在修補漏洞上更有效的同樣能力,也讓它在利用漏洞上同樣更有效。」這一說法凸顯出 Mythos 的雙面性:防禦與攻擊之間的界線,在此類模型面前幾近消弭。

Anthropic 拒絕公開公佈,啟動「玻璃翼計劃」限縮存取範圍

正因能力過於強大,Anthropic 決定不對外公開發佈 Mythos,採取限制存取的策略,僅開放給特定合作企業使用。該公司同步宣佈啟動名為「玻璃翼計劃」(Project Glasswing)的防禦性網路安全合作計劃,與 AWS、蘋果、思科、Google、摩根大通、微軟、輝達等逾 40 家企業夥伴協作,目標是在攻擊者出手之前,主動找出並修補關鍵軟體中的漏洞。

圖源:X/@AnthropicAI Anthropic 同步宣佈啟動名為「玻璃翼計劃」(Project Glasswing)的防禦性網路安全合作計劃

Anthropic 表示:「鑑於模型能力之強大,我們在發佈方式上採取審慎態度。我們正與一小群早期存取客戶合作測試該模型,並認為這是我們迄今構建的最具突破性的一代。」

該公司另透露,已就 Mythos 的攻防應用場景向政府官員進行簡報,並持續與美國網路安全暨基礎設施安全局(CISA)及 AI 標準創新中心保持協商。外界後續關注的焦點,則集中在監管機關將如何制定類似模型的發佈審查標準、金融機構應採取哪些具體防禦措施,以及國際監管機構之間的協調能否跟上技術演進的速度。

本篇內容由加密 Agent 匯總各方資訊生成、《加密城市》審稿與編輯,目前仍處於培訓階段,可能存在邏輯偏差或資訊誤差,內容僅供參考,請勿視為投資建議。

Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to Disclaimer.

Articoli correlati

South Korean AI Unicorn Upstage Shuts Down Hong Kong Office After 5 Years

South Korean AI company Upstage has closed its Hong Kong subsidiary after five years due to underperformance. The company is now focused on the U.S. and Japan, investing in its U.S. operations and expanding AI solutions through partnerships.

GateNews23m fa

汽油彈攻擊 Altman 住宅男子遭控謀殺未遂,隨身筆記列多名 AI 高管姓名地址

舊金山地檢對20歲男子Daniel Moreno-Gama提起起訴,因其曾向OpenAI執行長Sam Altman住宅投擲汽油彈,面臨2項謀殺未遂指控。FBI發現其手上有多名AI高管的住址名單,顯示潛在威脅範圍廣泛。案件引起高度關注,Altman呼籲各方降低對抗情緒。

動區BlockTempo26m fa

Hong Kong to Launch AI Research Institute in H2 2026, Targeting Healthcare, Legal, and Financial Sectors

Hong Kong's Chief Executive John Lee announced the launch of the Hong Kong Artificial Intelligence Research Institute in 2026 to boost AI in healthcare, legal, and financial sectors, supporting the national "AI+" initiative and enhancing innovation in the Greater Bay Area.

GateNews34m fa

SoftBank, NEC, Honda, Sony Launch Joint Venture to Develop Trillion-Parameter Physical AI Model

SoftBank, NEC, Honda, and Sony have formed a joint venture, Japan AI Foundation Model Development, to create a trillion-parameter AI model for industrial applications by 2030, with funding support from NEDO.

GateNews42m fa

Nvidia Launches AI Models for Quantum Error Correction on World Quantum Day

Nvidia launched the Ising suite, AI models for quantum calibration and error correction, achieving significant performance gains. The suite is utilized by various quantum firms and provides frameworks for easier AI integration, while Nvidia's AI vision extends to multiple industries, including partnerships with Samsung and Hyundai.

GateNews1h fa

Claw Intelligence Raises $3M Seed Round Led by Castrum Istanbul

Claw Intelligence has secured $3 million in seed funding to simplify user interactions with Web3 ecosystems. Its platform uses a Model Context Protocol to streamline data and includes features like a secure code execution sandbox and LLM-powered device control.

GateNews1h fa
Commento
0/400
Nessun commento