
? 人工智能公司Anthropic的首席信息安全官杰森·克林頓表示,具備記憶、崗位職責(zé)和企業(yè)憑證的AI“虛擬員工”可能僅需一年就能出現(xiàn)。但他向Axios指出,這類新型AI智能體將帶來(lái)獨(dú)特的網(wǎng)絡(luò)安全風(fēng)險(xiǎn)。
美國(guó)頂尖人工智能實(shí)驗(yàn)室之一的Anthropic預(yù)測(cè),由AI驅(qū)動(dòng)的虛擬員工最早將于明年進(jìn)入職場(chǎng)。
該公司首席信息安全官杰森·克林頓近日接受Axios采訪時(shí)表示,這些具備AI身份的虛擬員工將擁有自己的“記憶”、公司崗位職責(zé),甚至獨(dú)立的公司賬戶和密碼。
這類深度融入企業(yè)架構(gòu)的虛擬員工,標(biāo)志著對(duì)市場(chǎng)上現(xiàn)有AI智能體的重大升級(jí)。
當(dāng)前市面上的AI智能體更像是專用工具——設(shè)計(jì)用于處理明確指定的單一任務(wù),例如標(biāo)記釣魚(yú)攻擊或識(shí)別潛在安全威脅等。
相比之下,虛擬員工代表著一類全新的AI:這種智能系統(tǒng)具備更廣泛的職責(zé),更深度融入日常工作流程,并能自主決策。
但克林頓指出,這類AI智能體的出現(xiàn)將帶來(lái)大量網(wǎng)絡(luò)安全隱患,例如如何保障AI員工的賬戶安全,以及如何降低其行為失控的風(fēng)險(xiǎn)等。
他對(duì)媒體表示:“在這個(gè)領(lǐng)域,從安全角度看還有太多難題亟待解決?!彼€透露Anthropic已著手通過(guò)多種方法降低風(fēng)險(xiǎn),包括測(cè)試模型的抗網(wǎng)絡(luò)攻擊能力,以及監(jiān)控惡意行為者濫用系統(tǒng)帶來(lái)的安全隱患。
AI同事的崛起
公司日益熱衷于AI虛擬員工,因?yàn)橄啾热祟悊T工,虛擬員工能夠?qū)崿F(xiàn)更大幅度的降本增效。
Shopify首席執(zhí)行官托比·呂特克近期在內(nèi)部備忘錄中對(duì)員工明確表示,任何任務(wù)都應(yīng)優(yōu)先考慮AI工具,甚至早于申請(qǐng)?jiān)黾尤肆幹啤?/p>
他對(duì)員工表示,未來(lái)在批準(zhǔn)新錄用人員之前,必須證明為什么AI無(wú)法完成該工作。
Klarna首席執(zhí)行官塞巴斯蒂安·西米亞特科夫斯基更進(jìn)一步認(rèn)為,生成式AI已完全取代新增人力的需求。
他透露該公司一年前就停止了所有招聘,其AI助手承擔(dān)了約700名客服人員的工作量,將問(wèn)題解決時(shí)間從11分鐘縮短至2分鐘。
除成本與效率優(yōu)勢(shì)外,AI員工還可能開(kāi)創(chuàng)全新工作模式。AI在編程任務(wù)中表現(xiàn)尤為突出,谷歌(Google)、Meta等科技公司已將大量此類工作交由內(nèi)部AI系統(tǒng)處理。
虛擬員工可7×24小時(shí)運(yùn)轉(zhuǎn),并能高速處理海量數(shù)據(jù),有望顯著加速工作流程。
但AI的崛起引發(fā)了人們對(duì)大規(guī)模失業(yè)的擔(dān)憂。
研究表明,行政類和重復(fù)性強(qiáng)的崗位尤其容易受到先進(jìn)AI系統(tǒng)的沖擊。(財(cái)富中文網(wǎng))
譯者:劉進(jìn)龍
審校:汪皓
? 人工智能公司Anthropic的首席信息安全官杰森·克林頓表示,具備記憶、崗位職責(zé)和企業(yè)憑證的AI“虛擬員工”可能僅需一年就能出現(xiàn)。但他向Axios指出,這類新型AI智能體將帶來(lái)獨(dú)特的網(wǎng)絡(luò)安全風(fēng)險(xiǎn)。
美國(guó)頂尖人工智能實(shí)驗(yàn)室之一的Anthropic預(yù)測(cè),由AI驅(qū)動(dòng)的虛擬員工最早將于明年進(jìn)入職場(chǎng)。
該公司首席信息安全官杰森·克林頓近日接受Axios采訪時(shí)表示,這些具備AI身份的虛擬員工將擁有自己的“記憶”、公司崗位職責(zé),甚至獨(dú)立的公司賬戶和密碼。
這類深度融入企業(yè)架構(gòu)的虛擬員工,標(biāo)志著對(duì)市場(chǎng)上現(xiàn)有AI智能體的重大升級(jí)。
當(dāng)前市面上的AI智能體更像是專用工具——設(shè)計(jì)用于處理明確指定的單一任務(wù),例如標(biāo)記釣魚(yú)攻擊或識(shí)別潛在安全威脅等。
相比之下,虛擬員工代表著一類全新的AI:這種智能系統(tǒng)具備更廣泛的職責(zé),更深度融入日常工作流程,并能自主決策。
但克林頓指出,這類AI智能體的出現(xiàn)將帶來(lái)大量網(wǎng)絡(luò)安全隱患,例如如何保障AI員工的賬戶安全,以及如何降低其行為失控的風(fēng)險(xiǎn)等。
他對(duì)媒體表示:“在這個(gè)領(lǐng)域,從安全角度看還有太多難題亟待解決。”他還透露Anthropic已著手通過(guò)多種方法降低風(fēng)險(xiǎn),包括測(cè)試模型的抗網(wǎng)絡(luò)攻擊能力,以及監(jiān)控惡意行為者濫用系統(tǒng)帶來(lái)的安全隱患。
AI同事的崛起
公司日益熱衷于AI虛擬員工,因?yàn)橄啾热祟悊T工,虛擬員工能夠?qū)崿F(xiàn)更大幅度的降本增效。
Shopify首席執(zhí)行官托比·呂特克近期在內(nèi)部備忘錄中對(duì)員工明確表示,任何任務(wù)都應(yīng)優(yōu)先考慮AI工具,甚至早于申請(qǐng)?jiān)黾尤肆幹啤?/p>
他對(duì)員工表示,未來(lái)在批準(zhǔn)新錄用人員之前,必須證明為什么AI無(wú)法完成該工作。
Klarna首席執(zhí)行官塞巴斯蒂安·西米亞特科夫斯基更進(jìn)一步認(rèn)為,生成式AI已完全取代新增人力的需求。
他透露該公司一年前就停止了所有招聘,其AI助手承擔(dān)了約700名客服人員的工作量,將問(wèn)題解決時(shí)間從11分鐘縮短至2分鐘。
除成本與效率優(yōu)勢(shì)外,AI員工還可能開(kāi)創(chuàng)全新工作模式。AI在編程任務(wù)中表現(xiàn)尤為突出,谷歌(Google)、Meta等科技公司已將大量此類工作交由內(nèi)部AI系統(tǒng)處理。
虛擬員工可7×24小時(shí)運(yùn)轉(zhuǎn),并能高速處理海量數(shù)據(jù),有望顯著加速工作流程。
但AI的崛起引發(fā)了人們對(duì)大規(guī)模失業(yè)的擔(dān)憂。
研究表明,行政類和重復(fù)性強(qiáng)的崗位尤其容易受到先進(jìn)AI系統(tǒng)的沖擊。(財(cái)富中文網(wǎng))
譯者:劉進(jìn)龍
審校:汪皓
? Anthropic’s Chief Information Security Officer (CISO), Jason Clinton, has said AI “virtual employees”—complete with memories, roles, and corporate credentials—could be just one year away. However, Clinton told Axios that the new frontier of AI agents poses a unique set of cybersecurity risks.
Anthropic, one of the U.S.’s leading AI labs, anticipates that artificial intelligence-powered virtual employees could emerge in the workplace as early as next year.
In a recent interview with Axios, Jason Clinton, Anthropic’s Chief Information Security Officer (CISO), said the virtual employees with AI identities could have their own “memories,” roles in the company, and even their own corporate accounts and passwords.
Virtual employees integrated into a corporation would mark a significant step up from the AI agents currently on the market.
Today’s AI agents are like specialized tools—built to handle single, well-defined jobs like flagging phishing attempts or spotting potential security threats.
In contrast, virtual employees represent a whole new class of AI: intelligent systems with broader responsibilities, deeper integration into daily workflows, and the ability to make decisions independently.
However, Clinton said that the introduction of AI agents would bring with it a host of cybersecurity concerns, such as complications around securing the AI employee’s user accounts and mitigating the risk of them going rogue.
“In that world, there are so many problems that we haven’t solved yet from a security perspective that we need to solve,” Clinton told the outlet. He added that Anthropic was already working on mitigating the risks with methods such as testing its models to ensure they can withstand cyberattacks and monitor safety issues around the threat of abuse by malicious actors.
The rise of the AI co-worker
Companies have been increasingly enthusiastic about the prospect of AI virtual employees as they promise both cost savings and productivity gains at a scale that human workforces can’t match.
Shopify’s CEO Tobi Lütke made this explicit in a recent internal memo that told employees AI tools should be the first port of call for any task, even before requesting additional headcount.
He told employees that, moving forward, they must demonstrate why AI cannot accomplish a job before new hires are approved.
At Klarna, CEO Sebastian Siemiatkowski went further, arguing that generative AI has already supplanted the need for new hires.
He revealed that the company halted all hiring a year ago, as its AI assistants absorbed the workload of roughly 700 customer-service staff, cutting resolution times from 11 minutes to two.
Beyond cost and speed, AI employees could also unlock new ways of working. AI is particularly good at coding tasks, and several tech companies, including Google and Meta, are already outsourcing a sizable amount of this work to internal AI systems.
Virtual employees can operate 24/7 and handle vast amounts of data at speed, potentially speeding up workflows.
However, the rise of AI has caused concerns about mass job losses among employees.
Some jobs, often admin-heavy and repetitive, have been shown to be particularly exposed to advanced AI systems.