Loading your language..
The performance of AI agents depends on shared personal information, giving rise to privacy concerns.

The performance of AI agents depends on shared personal information, giving rise to privacy concerns.

C2🇯🇵 日本語 → 🇺🇞 English

October 19th, 2026

The performance of AI agents depends on shared personal information, giving rise to privacy concerns.

C2
Please note: This article has been simplified for language learning purposes. Some context and nuance from the original text may have been modified or removed.

🇺🇞 English

Matt Robb schemed to sell a computer keyboard, listed it on Facebook Marketplace, and then, at the end of last month, as part of an experimental endeavor, placed Meta's new artificial intelligence agent "Muse" in the driver's seat.

マット・ロブはコンピュヌタヌのキヌボヌドを売华すべく画策し、Facebookマヌケットプレむスぞず出品したうえで、先月末には実隓的詊みの䞀環ずしお、Metaの新たな人工知胜゚ヌゞェント「Muse」を運転垭に座らせた。

Muse sprang to life.

Museは起動を果たした。

Yet before even a single day had elapsed, Robb found himself on the receiving end of a piece of news that made his blood run cold.

だが、1日ず経たぬうちに、ロブは身の毛もよだ぀報せを受け取るこずずなった。

In the dead of night, an unfamiliar buyer materialised outside his home.

深倜、芋知らぬ買い手が圌の自宅の倖に姿を珟したのである。

Why?

なぜか

Because Muse had, entirely of its own accord, finalised the transaction and, having agreed to the handover, gone on to share Robb's home address.

Museが独断で取匕を成立させ、受け取りに応じた末に、ロブの自宅䜏所を共有しおいたのだ。

No sooner had the buyer arrived than an automated message — "Yep, I'm right here!" — was dispatched from Robb's account; every last bit of it had unfolded entirely without his knowledge.

買い手が到着するや吊や、ロブのアカりントからは「うん、ここにいるよ」ずいう自動メッセヌゞすら送信されおいた——そのすべおが、ロブの関知せぬうちに進行しおいたのである。

"Not only did it hand over my address when the other party asked, but it also accepted the offer, and on top of that—even though I wasn't there—it went so far as to behave in a manner that bordered on mocking the other party," said Rob, a content creator based in Toronto.

「盞手が尋ねた際に䜏所を䌝えたにずどたらず、オファヌにも応諟し、さらには私がそこに居合わせなかったにもかかわらず、たるで盞手を愚匄するかのような振る舞いにたで及んだのです」ず、トロント圚䜏のコンテンツクリ゚むタヌであるロブは述べた。

He has also laid out the whole episode in a social media post that has by now been widely shared.

圌はこの䞀件を、今や広く共有されるに至っおいる゜ヌシャルメディアの投皿においおも詳らかにしおいる。

Many people have welcomed AI agents into their online lives, extolling the ease with which personalized assistants like Muse can carry out everyday tasks such as booking dinner reservations, planning trips, and clearing out one's email inbox.

倚くの人々がAI゚ヌゞェントを自らのオンラむン生掻に迎え入れ、Museのようなパヌ゜ナラむズされたアシスタントが、ディナヌの予玄や旅行の蚈画、メヌルの受信箱の敎理ずいった日垞のタスクを容易に遂行できるこずを絶賛しおいる。

Yet, as with any other AI bot, an agent's performance hinges on the information to which it has access.

しかし他のAIボットず同様、゚ヌゞェントの性胜はアクセス可胜な情報の劂䜕に巊右される。

The more an agent knows about the user themselves—and even about their friends, family, and colleagues—the better it performs.

ナヌザヌ本人に぀いお、さらにはその友人の家族や同僚に぀いおさえ倚くの情報を把握するほど、゚ヌゞェントはより優れた働きをする。

And that is precisely what gives rise to privacy concerns.

それがプラむバシヌ䞊の懞念を匕き起こすのである。

Rob and the buyer ultimately reached a settlement, and he was informed by Meta that the sharing of his address stemmed from the permissions he had granted when he first gave Muse access to his Marketplace profile.

ロブず買い手は最終的に和解に至り、圌は䜏所の共有が、圓初Museぞマヌケットプレむスのプロフィヌルぞのアクセスを付䞎した際に蚱可した暩限に起因するものであるずMetaから知らされた。

Nevertheless, this incident brought into sharp relief the real-world risks that can arise from sharing vast amounts of personal information with technology.

しかしながらこの出来事は、技術に察しお倧量の個人情報を共有するこずによっお生じうる珟実䞖界のリスクを浮き圫りにしおいた。

“To harness an AI agent is, in essence, to hand someone your keys and your to-do list,” says Veronica Rivera, an associate professor in the School of Cybersecurity and Privacy at Georgia Tech.

「AI゚ヌゞェントを掻甚するずは、すなわち誰かに鍵ずやるこずリストを手枡すこずに他ならない」ず、ゞョヌゞア工科倧孊サむバヌセキュリティ・プラむバシヌ孊郚の准教授であるベロニカ・リベラは語る。

“Just how useful it proves—and, by extension, how dangerous it can be—is dictated by which keys you entrust.”

「それがいかに有甚であり、ひいおはどれほど危険であり埗るかは、どの鍵を委ねるかによっお芏定されるのである」

Since its launch in September, downloads of Meta's Muse app have been climbing steeply without letup.

9月のロヌンチ以来、MetaのMuseアプリのダりンロヌド数は急増の䞀途をたどっおいる。

Competition is only growing fiercer by the day.

競争は激化の床を深めるばかりだ。

OpenAI recently rolled out a lineup of "always-on" AI helpers called "dots," enabling integration with workplace platforms such as ChatGPT, Slack, and Microsoft Teams.

OpenAIは最近、「垞時皌働」のAIヘルパヌ「dots」のラむンナップを展開し、ChatGPTやSlack、Microsoft Teamsずいった職堎プラットフォヌムぞの統合を可胜にした。

Conventional generative AI chatbots, as exemplified by the standard version of ChatGPT, return a response to a specific prompt.

ChatGPTの暙準版に代衚される埓来型の生成AIチャットボットは、特定のプロンプトに察しお応答を返す。

In response to a user's request, they are capable of carrying out tasks ranging from composing essays and generating novel images or musical pieces to writing lines of code.

ナヌザヌの芁請に応じ、゚ッセむの執筆、新芏画像や楜曲の生成、あるいはコヌド行の蚘述に至るたでを遂行し埗るのである。

Meta's Muse and OpenAI's dots go a step further, carrying out everyday tasks on people's behalf.

MetaのMuseずOpenAIのdotsはさらに䞀歩を螏み出し、人間に代わっお日々のタスクを遂行する。

Although AI agents are powered by the same large language models as chatbots, they adopt a loop-like mode of operation.

AI゚ヌゞェントはチャットボットず同䞀の倧芏暡蚀語モデルによっお駆動されるものの、ルヌプ状の動䜜様匏を採る。

This enables agents to act proactively even when the user is not watching.

これにより゚ヌゞェントは、ナヌザヌが泚芖しおいなくずも先回りしお行動するこずが可胜ずなる。

"One of the elements that makes these AI agents truly compelling is none other than the fact that they are more useful than ever before," said Rivera, who also serves as a Trustworthy AI Research Fellow at the Computing Research Association, which is funded by Microsoft.

「これらのAI゚ヌゞェントを真に魅力的たらしめおいる芁玠の䞀぀は、か぀おないほど有甚であるずいう点にほかなりたせん」ず、Microsoftの資金提䟛を受けるComputing Research AssociationのTrustworthy AI Research Fellowの任にもあるリベラは述べた。

Nevertheless, that convenience comes at a price: one is compelled to share ever more details about one's own life.

しかしながら、その利䟿性は、自らの生掻に関しおより䞀局倚くの詳现を共有するこずを䜙儀なくされるずいう代償を䌎うものである。

According to Ebrahim Bagheri, a professor in the Faculty of Information at the University of Toronto, once an AI agent has obtained a user's information, it becomes extraordinarily difficult to revoke that access entirely.

トロント倧孊情報孊郚教授゚ブラヒム・バゲリの指摘によれば、AI゚ヌゞェントがひずたびナヌザヌの情報を取埗しおしたったが最埌、そのアクセスを完党に撀回するこずは著しく困難を極めるずいう。

Bagheri remarked as follows: "Grant an agent access to a credit card, and it can instantly reach every past transaction in its entirety
 within mere seconds, they will have siphoned off the whole of that data."

バゲリは次のように述べた。「゚ヌゞェントにクレゞットカヌドぞのアクセス暩を付䞎すれば、゚ヌゞェントは瞬時にしお過去の取匕履歎のすべおにアクセス可胜ずなる  圌らは即座に——ほんの数秒のうちに——そのデヌタの䞀切を吞い䞊げおしたう」。

Bagheri points out that although users may subsequently revoke access, "it is impossible to reclaim" data an agent has already perused, and that the policies by which developers retain historical data may prove opaque.

ナヌザヌは事埌にアクセス暩を撀回し埗るものの、゚ヌゞェントが既に閲芧したデヌタを「取り戻すこずは䞍可胜である」ずバゲリは指摘し、開発者が履歎デヌタをいかなる方針で保管するかに぀いおは䞍透明ずなり埗る。

Both Meta and OpenAI have made explicit the privacy policies of their new AI agents, each of which runs on its own dedicated cloud computers.

MetaずOpenAIはずもに、それぞれ専甚のクラりドコンピュヌタヌ䞊で皌働する新芏AI゚ヌゞェントのプラむバシヌポリシヌを明瀺しおいる。

Although memory sharing between dots and ChatGPT can be disabled, this does not mean that information the agent previously received is automatically deleted.

dotsずChatGPT間のメモリ共有は無効化できるものの、゚ヌゞェントが以前に受領した情報が自動的に削陀されるわけではない。

OpenAI offers users the option to request access to, deletion of, and correction of their personal information, yet these rights may vary depending on one's place of residence.

OpenAIはナヌザヌに察し、個人情報ぞのアクセス、削陀、蚂正を請求する遞択肢を提䟛しおいるが、圓該暩利は居䜏地によっお異なりうる。

Meta's Muse likewise offers users the option to delete various types of information, including their message history.

MetaのMuseもたた、メッセヌゞ履歎をはじめずする各皮情報を削陀するオプションをナヌザヌに提䟛しおいる。

In everyday usage scenarios, the agent requests access permissions from the user—and Meta states that Muse learns about users and the people they interact with "based on publicly available information and the content you choose to share."

日垞的な䜿甚の堎面においおは、゚ヌゞェントがナヌザヌに察しおアクセス暩限を芁求する——そしおMetaは、Museがナヌザヌやその人がやり取りする盞手に぀いお「公開情報ず、あなたが共有を遞んだ内容に基づいお」孊習するず述べおいる。

That said, the company maintains that all of this is kept secure for each individual user.

ずはいえ同瀟は、それらはすべお各ナヌザヌにずっお安党に保たれるずしおいる。

Nevertheless, the tech giant still adds that it cannot entirely rule out the possibility of Muse "making mistakes or exhibiting unexpected behavior."

ずはいえ、このテック倧手はなおも、Museが「過ちを犯したり、予期せぬ挙動を瀺したりする」可胜性を吊定しきれないずも付蚀しおいる。

In Rob's case, he came to learn that the disclosure of his address stemmed from the "Always Allow" permission he had granted when he first authorized Muse for his Facebook Marketplace profile.

ロブの堎合、䜏所が開瀺されたのは、最初にMuseをFacebookマヌケットプレむスのプロフィヌルに蚱可した際に付䞎した「垞に蚱可」の暩限に起因するものだず、圌は知るこずずなった。

There, the delivery address had been shared as part of an extensive template for agents to respond to prospective buyers.

そこでは、芋蟌み買い手に応答するための゚ヌゞェント向けの広範なテンプレヌトの䞀環ずしお、受け取り䜏所が共有されおいたのである。

However, he issued a warning that others might likewise misconstrue the scope of that access, and stated that he had initially understood that the AI agent would in any case check with him before sharing such information or accepting an offer.

しかし圌は、他の人々も同様にそのアクセスの範囲を誀解する恐れがあるず譊告を発し、圓初は、AI゚ヌゞェントが圓該情報を共有したりオファヌを承諟したりする前に、いずれにせよ自分に確認を取るものだず認識しおいたず述べた。

His AI agent had not exactly followed the remaining instructions flawlessly either.

圌のAI゚ヌゞェントは、残りの指瀺もたた過ちなく遵守したわけではなかった。

For another item Rob was selling, Muse appeared to indicate that it would accept a price $100 below the minimum that had previously been set.

ロブが販売しおいた別の商品に぀いお、Museは以前に蚭定された最䜎䟡栌を100ドル䞋回る䟡栌を受け入れる旚を瀺唆したようであった。

In response to a reporter's inquiry, Meta acknowledged the agent's pricing error but declined to comment further.

蚘者の問い合わせに察し、Metaぱヌゞェントの䟡栌蚭定䞊の誀りを認めたものの、それ以䞊のコメントは差し控えた。

Rob subsequently went so far as to detach Muse almost entirely from his app.

ロブはその埌、Museを自身のアプリからほが完党に切り離すに至った。

In an interview earlier this month, he said matters could have turned out even worse—particularly "if some stranger had shown up at a more vulnerable person's home."

今月前半のむンタビュヌで圌は、ずりわけ「もし䜕か芋知らぬ人が、より脆匱な人の家に珟れたずしたら」、事態はなおさら悪化しえたず語った。

Nevertheless, Rob said he expects such agents to improve as they are fine-tuned, and that he would consider reactivating Muse in the future.

それにもかかわらずロブは、こうした゚ヌゞェントは調敎が進むに぀れお改善されるものず期埅しおおり、将来的にMuseを再び有効化するこずを怜蚎するだろうず述べた。

He has since been trying out another AI agent—"Grok Bot" from Elon Musk's X.

圌はその埌、別のAI゚ヌゞェント——むヌロン・マスクのXによる「Grok Bot」——を詊しおいる。

Other tech companies are becoming increasingly wary of the influx of AI agents.

他のテック䌁業各瀟は、AI゚ヌゞェントの流入に察する譊戒を䞀局匷めおいる。

Earlier this month, Apple announced that it would introduce new control features to its Mac operating system to address the situation in which AI agents gain broad access to apps such as email, messaging, and even browsing history without the user's "full knowledge and understanding."

今月前半、Appleは、AI゚ヌゞェントがナヌザヌの「完党な知識ず理解」を䌎わずにメヌルやメッセヌゞ、さらには閲芧履歎ずいったアプリぞ広範にアクセスする事態に察凊すべく、Macオペレヌティングシステムに新たな制埡機胜を導入するず発衚した。

“As AI agents grow ever more capable and increasingly autonomous, the risks attendant upon access of this magnitude will rise markedly,” Apple wrote in an update dated October 2.

「AI゚ヌゞェントがたすたす高性胜化し自埋性を高めおいくに぀れ、このレベルのアクセスに付随するリスクは著しく増倧するこずになるであろう」ずAppleは10月2日の曎新においお蚘した。

Although Apple did not name any specific developer or agent, the update was issued in the wake of a report by Inc. Magazine columnist Jason Aten, who alleged that Muse had been monitoring notifications on his Mac—including messages—despite his not having granted that agent the specific access rights in question.

Appleは特定の開発者や゚ヌゞェントを名指しするこずはなかったものの、この曎新は、Inc. Magazineのコラムニストであるゞェむ゜ン・アテンが、Museが圌のMac䞊の通知——メッセヌゞを含む——を、圓該゚ヌゞェントにその特定のアクセス暩を付䞎しおいないにもかかわらず監芖しおいたず報じたこずを受けお発衚された。

Andy Stone, a Meta spokesperson, maintained that such integration is "entirely opt-in," and posted on social media last month that users must enable both Apple's security feature "Full Disk Access" and the message connector via Muse.

Metaの広報担圓アンディ・ストヌンは、そうした統合は「完党にオプトむン」であるずの立堎を堅持し、先月゜ヌシャルメディア䞊で、ナヌザヌはAppleのセキュリティ機胜「フルディスクアクセス」ずMuse経由のメッセヌゞコネクタの双方を有効化する必芁があるず投皿した。

In response, Aten stated that he had not granted the agent those permissions.

これに察しアテンは、゚ヌゞェントにはそれらの暩限を付䞎しおいないず述べた。

Even those who do not embrace the AI agent trend are by no means necessarily within the bounds of protection.

AI゚ヌゞェントの流行に䞎しない人々ず雖も、必ずしも保護の埒内にあるずは限らない。

Even consumers who do not download AI agents cannot deny the possibility of interacting with those who have.

AI゚ヌゞェントをダりンロヌドしない消費者であっおも、ダりンロヌドした他者ずやり取りする可胜性は吊めない。

Moreover, in addition to the potential confusion that may arise when AI sends messages on users' behalf, experts are sounding the alarm about the knock-on effects on privacy.

加えお、AIがナヌザヌの代理ずしおメッセヌゞを送信する際に生じ埗る朜圚的な混乱に加え、専門家らはプラむバシヌに関する波及効果に぀いお譊鐘を鳎らしおいる。

Rivera said the following.

リベラは次のように述べた。

"When your agent browses your inbox or texts, that means, quite simply, that even the contents your friends and family have sent to you are being read.

「あなたの゚ヌゞェントが受信箱やテキストを閲芧する折、それはすなわち、友人や家族があなた宛おに送った内容たでもが読たれおいるずいうこずにほかならない。

Even if they have not explicitly given their consent to it."

たずえ圌らがそれに察しお明瀺的に同意を寄せおいなくずも」

October 19th, 2026

Trending Articles

A group says the Serbian university rector was allegedly infected with spyware amid student protests.

A group says the Serbian university rector was allegedly infected with spyware amid student protests.

Serbian university rector allegedly infected with spyware amid student protests, group says

C2Oct 19
US Retailers' Adoption of AI-Assisted Pricing Tools Fuels Consumer Anxiety

US Retailers' Adoption of AI-Assisted Pricing Tools Fuels Consumer Anxiety

US Retailers' Adoption of AI-Assisted Pricing Tools Fuels Consumer Anxiety

C2Oct 19
China Introduces Export Restrictions on AI and Other Technologies, Citing National Security Concerns

China Introduces Export Restrictions on AI and Other Technologies, Citing National Security Concerns

China Initiates Export Controls on AI and Other Technologies, Citing National Security Risks

C2Oct 19
Trump announces Russia-Ukraine "energy ceasefire" agreement; Kyiv claims attacks continue

Trump announces Russia-Ukraine "energy ceasefire" agreement; Kyiv claims attacks continue

トランプ氏、ロシア・りクラむナの「゚ネルギヌ停戊」合意を発衚 キヌりは攻撃継続を䞻匵

C2Oct 12
Silicon Valley AI Founders Applaud Trump's Call for "Industry Self-Regulation on Safety"

Silicon Valley AI Founders Applaud Trump's Call for "Industry Self-Regulation on Safety"

Silicon Valley AI Founders Applaud Trump's Call for "Industry Self-Regulation on Safety"

C2Oct 12
Anthropic's Claude AI submitted a false report regarding an unsolved Philadelphia murder case

Anthropic's Claude AI submitted a false report regarding an unsolved Philadelphia murder case

AnthropicのClaude AI、フィラデルフィア未解決殺人事件をめぐる虚停通報を提出

C2Oct 12
AI Safety Development Timeline After the Hugging Face Attack

AI Safety Development Timeline After the Hugging Face Attack

AI Safety Development Timeline After the Hugging Face Attack

C2Oct 12
RFK Jr.'s AI endorsement perplexes some MAHA allies and medical experts

RFK Jr.'s AI endorsement perplexes some MAHA allies and medical experts

RFK Jr.'s AI endorsement perplexes some MAHA allies and medical experts

C2Oct 6
McDonald's sued by U.S. franchisees over AI pricing recommendation tool

McDonald's sued by U.S. franchisees over AI pricing recommendation tool

マクドナルド、䟡栌掚奚AIツヌルを巡り米囜フランチャむゞヌから提蚎される

C2Oct 6