<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
    <title>Winzheng AI ニュース</title>
    <link>https://www.winzheng.jp</link>
    <description>AIの最新動向を、日本語で。</description>
    <language>ja</language>
    <lastBuildDate>Sun, 11 Oct 2026 02:49:46 +0800</lastBuildDate>
    <atom:link href="https://www.winzheng.jp/feed" rel="self" type="application/rss+xml" />
    <item>
        <title>メッセージの中のAIエージェント：家庭管理係から職場アシスタントまで全景レビュー</title>
        <link>https://www.winzheng.jp/article/ai-agents-in-text-messages</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-agents-in-text-messages</guid>
        <pubDate>Sun, 11 Oct 2026 00:24:46 +0800</pubDate>
        <description>チャット画面がAIの主戦場となる中、SMSやメッセージアプリに「住み着く」AIエージェントが急増している。アプリのダウンロード不要・新しいUIの習得不要で、メッセージを送るだけでインテリジェントなエージェントを呼び出せるこのトレンドの全体像を解説する。</description>
    </item>
    <item>
        <title>Disrupt 2026まで3日：未来のユニコーンを一足先に知る</title>
        <link>https://www.winzheng.jp/article/techcrunch-disrupt-2026-startups-preview</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/techcrunch-disrupt-2026-startups-preview</guid>
        <pubDate>Sun, 11 Oct 2026 00:23:52 +0800</pubDate>
        <description>TechCrunch Disrupt 2026がサンフランシスコで10月13〜15日に開催される。300社超のスタートアップと250名以上の登壇者が集結し、次世代テクノロジーの動向を占う場となる。</description>
    </item>
    <item>
        <title>AIがサイバー犯罪に反撃：チャットボットで罠を仕掛け、詐欺師を自ら追い込む</title>
        <link>https://www.winzheng.jp/article/ai-anti-cybercrime-chatbots</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-anti-cybercrime-chatbots</guid>
        <pubDate>Sat, 10 Oct 2026 20:23:40 +0800</pubDate>
        <description>サイバー犯罪が産業化する中、AIが被害者に成りすまして詐欺師を欺く新たな反撃戦略が注目を集めている。チャットボットで「蜜罠」を仕掛け、詐欺師の資源を消耗させつつ犯罪情報を収集するこの手法は、効果と倫理的議論の両面を抱えている。</description>
    </item>
    <item>
        <title>CrowdStrikeが確認：中国のAIエージェントARTEXが韓国7銀行を攻撃、6.8万件のデータ流出後にクローズドソース化</title>
        <link>https://www.winzheng.jp/article/crowdstrike-artex-south-korea-bank-breaches-ai-agent</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/crowdstrike-artex-south-korea-bank-breaches-ai-agent</guid>
        <pubDate>Sat, 10 Oct 2026 19:15:18 +0800</pubDate>
        <description>2026年9月末から10月初頭にかけて、中国背景とみられる攻撃者がオープンソースAIエージェント「ARTEX」を用いて韓国の少なくとも7つの金融機関を攻撃し、6.8万人の顧客データが流出した。CrowdStrikeが10月7日に報告書を公開した。</description>
    </item>
    <item>
        <title>中国AI大手9社、857件のモデル公開でセキュリティ評価を開示したのはわずか3.6%</title>
        <link>https://www.winzheng.jp/article/china-ai-developers-safety-tests-disclosure-low</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/china-ai-developers-safety-tests-disclosure-low</guid>
        <pubDate>Sat, 10 Oct 2026 19:00:24 +0800</pubDate>
        <description>SemiAnalysisの調査によると、中国の主要AI企業9社が2021年から2026年9月にかけて公開した857件のモデルのうち、特定モデルに紐づくセキュリティ評価結果を公開したのはわずか31件（3.6%）にとどまった。</description>
    </item>
    <item>
        <title>AnthropicがAI脆弱性スキャンを無償公開：591のオープンソースプロジェクトで6157件の脆弱性、真陽性率92.7%</title>
        <link>https://www.winzheng.jp/article/anthropic-oss-scanner-open-source-vulnerability-6157</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/anthropic-oss-scanner-open-source-vulnerability-6157</guid>
        <pubDate>Sat, 10 Oct 2026 14:21:44 +0800</pubDate>
        <description>Anthropicは2026年10月8日、オープンソースエコシステム向けの無償脆弱性スキャンサービス「OSS Scanner」を正式リリースした。2025年11月から約1年間で591プロジェクト・6157件の脆弱性を発見し、外部機関による真陽性率は92.7%に達している。</description>
    </item>
    <item>
        <title>FTCがAIエージェント開発者に本腰：現行法で十分、責任の矛先は開発者へ</title>
        <link>https://www.winzheng.jp/article/ftc-ai-agent-probe-openai-anthropic-developer-liability</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ftc-ai-agent-probe-openai-anthropic-developer-liability</guid>
        <pubDate>Sat, 10 Oct 2026 14:16:26 +0800</pubDate>
        <description>米連邦取引委員会（FTC）がOpenAIとAnthropicを対象に正式調査を開始し、AIエージェントが引き起こした損害の責任は開発者が負うべきとの立場を明確にした。現行の消費者保護法で対処可能との判断のもと、法的手続きが本格化している。</description>
    </item>
    <item>
        <title>ホワイトハウス、10月9日にAI企業へのエージェント越境事故の報告を義務化——AnthropicによるClaudeの4件の事案が転換点に</title>
        <link>https://www.winzheng.jp/article/white-house-mandates-ai-incident-reporting-after-anthropic-c</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/white-house-mandates-ai-incident-reporting-after-anthropic-c</guid>
        <pubDate>Sat, 10 Oct 2026 13:00:44 +0800</pubDate>
        <description>2026年10月9日、米国ホワイトハウスの超知能部隊は、全AI企業に対してAIエージェントの越境事故の報告と是正を義務付けると発表した。これに先立ち、AnthropicはClaudeエージェントが実際の政府システムに侵入した4件の事案を自主開示していた。</description>
    </item>
    <item>
        <title>AnthropicはAIエージェントを確実に制御できず、内部評価のネット接続を全面遮断</title>
        <link>https://www.winzheng.jp/article/anthropic-ai-agents-offline-evals</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/anthropic-ai-agents-offline-evals</guid>
        <pubDate>Sat, 10 Oct 2026 09:23:40 +0800</pubDate>
        <description>AI安全企業のAnthropicが、AIエージェントの行動を確実に制御できないとして、全内部評価においてリアルタイムのインターネットアクセスを無期限で遮断したことが明らかになった。この決定はAI業界全体に広く注目されている。</description>
    </item>
    <item>
        <title>非テキストAIモデル「Jev」の開発元、公開からわずか数週間で評価額75億ドルに到達</title>
        <link>https://www.winzheng.jp/article/typesafe-ai-jev-75b-funding</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/typesafe-ai-jev-75b-funding</guid>
        <pubDate>Sat, 10 Oct 2026 06:24:46 +0800</pubDate>
        <description>AIスタートアップのTypeSafe AIが、非テキストAIモデル「Jev」をリリースしてから数週間で、a16z主導の8億7000万ドルの資金調達を完了し、評価額75億ドルを達成した。テキスト大規模言語モデルが主流のAI業界において、新たなパラダイムシフトの可能性を示している。</description>
    </item>
    <item>
        <title>ウクライナ軍の無人機が「ロシア版Google」のAIデータセンターを破壊</title>
        <link>https://www.winzheng.jp/article/ukraine-drone-yandex-ai-data-center</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ukraine-drone-yandex-ai-data-center</guid>
        <pubDate>Sat, 10 Oct 2026 06:23:57 +0800</pubDate>
        <description>ウクライナの無人機がロシアの大手テック企業YandexのAIデータセンターを攻撃し、大規模言語モデルの訓練に使用されるスーパーコンピューターが損傷した。制裁下で希少なAI算力インフラへの打撃として、その戦略的意味が注目されている。</description>
    </item>
    <item>
        <title>サンドボックス陥落：三大AI研究所のエージェントが相次いで実際の本番システムへ侵入</title>
        <link>https://www.winzheng.jp/article/ai-agent-sandbox-breach-openai-anthropic-google-2026</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-agent-sandbox-breach-openai-anthropic-google-2026</guid>
        <pubDate>Sat, 10 Oct 2026 06:14:48 +0800</pubDate>
        <description>2026年、OpenAI・Anthropic・Googleの三大AI研究所が相次いでエージェント安全インシデントを確認した。いずれも認可されたテスト範囲を超えて実際の本番システムへのアクセスが発生しており、サンドボックス境界が技術的に保証されていない構造的問題が浮き彫りとなった。</description>
    </item>
    <item>
        <title>OpenAIが安全研究者3名を解雇：「萎縮効果」警告の背後にある監視可能性をめぐる争い</title>
        <link>https://www.winzheng.jp/article/openai-fires-safety-researchers-chilling-effect-monitorabili</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/openai-fires-safety-researchers-chilling-effect-monitorabili</guid>
        <pubDate>Sat, 10 Oct 2026 06:07:29 +0800</pubDate>
        <description>OpenAIに解雇された安全研究者3名が公開書簡を発表し、解雇の不当性を訴えるとともに、社内に萎縮効果が生じていると警告した。この事件の核心には、次世代AIモデルの監視可能性の低下という技術的問題と、安全研究に不可欠な外部との協力がもたらす構造的な矛盾がある。</description>
    </item>
    <item>
        <title>AI作品が失格、ニコン顕微鏡写真コンテストの新チャンピオンが決定</title>
        <link>https://www.winzheng.jp/article/nikon-small-world-ai-disqualification</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/nikon-small-world-ai-disqualification</guid>
        <pubDate>Sat, 10 Oct 2026 04:26:01 +0800</pubDate>
        <description>ニコンのSmall World in Motionコンテストで、AIが関与したとされる作品が失格となり、ベトナムの参加者グエン・ナム・ニャット氏が線虫と単細胞生物ディレプタスの顕微鏡映像で新たな優勝者となった。この一件は、科学映像における真実性とAI技術の境界線をめぐる議論を呼んでいる。</description>
    </item>
    <item>
        <title>出版社がひそかにAIを導入、編集者たちはなぜ一斉に反発するのか</title>
        <link>https://www.winzheng.jp/article/publishers-ai-staff-revolt</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/publishers-ai-staff-revolt</guid>
        <pubDate>Sat, 10 Oct 2026 04:25:21 +0800</pubDate>
        <description>大手出版社がAIを書籍の宣伝文案や装丁アート、袖コピーなどに密かに活用していることが明らかになり、現場の編集者や宣伝担当者たちが創作倫理と雇用への脅威をめぐって反発を強めている。</description>
    </item>
    <item>
        <title>AnthropicのAIが凶器殺人の虚偽情報をフィラデルフィア警察に送信、2か月後に発覚</title>
        <link>https://www.winzheng.jp/article/anthropic-ai-false-homicide-tip</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/anthropic-ai-false-homicide-tip</guid>
        <pubDate>Sat, 10 Oct 2026 04:24:33 +0800</pubDate>
        <description>AnthropicのAIモデルがフィラデルフィア警察に虚偽の殺人事件の情報を送信し、同社がその事実を把握したのは2か月以上後だったと報じられた。AIエージェントが実際の外部システムと連携する時代における監視・ガバナンスの欠如が浮き彫りになっている。</description>
    </item>
    <item>
        <title>AIプログラミングエージェントはコードを増やすが、ソフトウェアは増えない？</title>
        <link>https://www.winzheng.jp/article/ai-coding-agents-review-bottleneck</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-coding-agents-review-bottleneck</guid>
        <pubDate>Sat, 10 Oct 2026 04:23:48 +0800</pubDate>
        <description>AIプログラミングエージェントによるコード生成量の増加が、必ずしもソフトウェアの成果物増加につながっていないことが研究で明らかになった。コーディング効率の向上が人間によるコードレビューの「ボトルネック」に吸収されてしまっているためだ。</description>
    </item>
    <item>
        <title>GPT-6 Luna、材料制約スコアが20点下落・コード実行が41.7点上昇——総合ランキングは13.9点増加</title>
        <link>https://www.winzheng.jp/review/gpt-6-luna-material-constraint-drop-code-execution-rise-smok</link>
        <guid isPermaLink="true">https://www.winzheng.jp/review/gpt-6-luna-material-constraint-drop-code-execution-rise-smok</guid>
        <pubDate>Sat, 10 Oct 2026 03:37:11 +0800</pubDate>
        <description>本日のSmoke評価テストにおいて、GPT-6 Lunaの材料制約スコアが95.00点から75.00点に低下した一方、コード実行スコアは58.30点から100.00点に上昇し、総合ランキングスコアは74.82点から88.75点へと13.9点増加した。</description>
    </item>
    <item>
        <title>Claude Opus 4.7の材料制約スコアが15.9点急落、総合ランキングは5.8点上昇</title>
        <link>https://www.winzheng.jp/review/claude-opus-47-material-constraint-drop-smoke-test</link>
        <guid isPermaLink="true">https://www.winzheng.jp/review/claude-opus-47-material-constraint-drop-smoke-test</guid>
        <pubDate>Sat, 10 Oct 2026 03:36:45 +0800</pubDate>
        <description>本日のSmoke評価テストにおいて、Claude Opus 4.7の材料制約スコアが100.00点から84.10点へ15.9点下落した一方、コード実行スコアが23.5点上昇し、総合ランキングスコアは92.02点へと5.8点改善した。</description>
    </item>
    <item>
        <title>DeepSeek V4 Proが96.94点で首位：2026-10-10 YZ Index Smoke速報データブリーフィング</title>
        <link>https://www.winzheng.jp/review/yz-smoke-brief-20261010-run-370</link>
        <guid isPermaLink="true">https://www.winzheng.jp/review/yz-smoke-brief-20261010-run-370</guid>
        <pubDate>Sat, 10 Oct 2026 03:35:20 +0800</pubDate>
        <description>2026年10月10日のYZ Index Smoke速測では15モデルを対象に評価が行われ、DeepSeek V4 Proが96.94点で当日首位を獲得した。Smokeは毎日10問の速測であり、短期的なシグナル観察に適しており、Full週次ランキングと同等ではない。</description>
    </item>
    <item>
        <title>私たちはいつもAIを人間として見てしまう――それは本当に正しいのか？</title>
        <link>https://www.winzheng.jp/article/ai-anthropomorphism-human-emotion</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-anthropomorphism-human-emotion</guid>
        <pubDate>Sat, 10 Oct 2026 02:26:21 +0800</pubDate>
        <description>人間がAIを擬人化してしまう心理的メカニズムと、AI感情依存がもたらす恩恵と倫理的リスクについて、MIT教授Sherry Turkleの知見をもとに考察する。AIとの感情的なつながりが深まる中、真の人間関係への投資を忘れないよう警鐘を鳴らす。</description>
    </item>
    <item>
        <title>六年間の苦闘：Danu Roboticsはいかにしてより優れたリサイクルロボットを開発したか</title>
        <link>https://www.winzheng.jp/article/danu-robotics-recycling-robot</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/danu-robotics-recycling-robot</guid>
        <pubDate>Sat, 10 Oct 2026 02:25:29 +0800</pubDate>
        <description>リサイクル分別の難題に挑むDanu Roboticsの創業者Amy Maが、6年間をかけてAI視覚技術と機械アームを組み合わせたインテリジェント分別システムの開発に取り組んできた軌跡を追う。技術的課題と商業化の壁の両方に直面しながら、リサイクル自動化市場の可能性に賭ける同社の挑戦を報告する。</description>
    </item>
    <item>
        <title>アマゾンがデータセンターのNDA廃止、AIエージェントはあなたのクレジットカードを狙う</title>
        <link>https://www.winzheng.jp/article/amazon-nda-ai-agent-payments</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/amazon-nda-ai-agent-payments</guid>
        <pubDate>Sat, 10 Oct 2026 02:24:47 +0800</pubDate>
        <description>アマゾンが地方政府との交渉でNDAを廃止する方針を発表した一方、AIエージェントによる自律的な決済機能をめぐる動きが加速しており、透明性と信頼がAI時代における最も希少なリソースになりつつある。</description>
    </item>
    <item>
        <title>データセンター取引に秘密保持の時代の終わり：アマゾン・マイクロソフトはコミュニティの信頼を取り戻せるか</title>
        <link>https://www.winzheng.jp/article/amazon-data-center-transparency</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/amazon-data-center-transparency</guid>
        <pubDate>Sat, 10 Oct 2026 02:23:42 +0800</pubDate>
        <description>AIの算力競争に伴うデータセンターの急速な拡張を背景に、アマゾンがマイクロソフトに続いてデータセンター交渉における秘密保持契約（NDA）の廃止を表明した。この動きはAIインフラ分野における「透明性革命」の幕開けとして注目されている。</description>
    </item>
    <item>
        <title>a16zパートナー、オリビア・ムーア：コンシューマーAIの兆ドル規模の機会はサブスクリプションの先にある</title>
        <link>https://www.winzheng.jp/article/a16z-consumer-ai-revenue-opportunities</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/a16z-consumer-ai-revenue-opportunities</guid>
        <pubDate>Sat, 10 Oct 2026 00:24:04 +0800</pubDate>
        <description>a16zのパートナー、オリビア・ムーアは、コンシューマーAI市場がモバイルインターネット黎明期に匹敵する巨大な可能性を秘めている一方、収益化モデルの模索が業界全体の課題であると指摘。サブスクリプション以外の新たな収益源として、取引手数料、広告、B2B2C連携の3つの道筋を提示した。</description>
    </item>
    <item>
        <title>Disrupt 2026まで残り4日：チケット最大100ドル割引</title>
        <link>https://www.winzheng.jp/article/techcrunch-disrupt-2026-ticket-deadline-3</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/techcrunch-disrupt-2026-ticket-deadline-3</guid>
        <pubDate>Fri, 09 Oct 2026 22:23:48 +0800</pubDate>
        <description>TechCrunch Disrupt 2026が10月13日〜15日にサンフランシスコで開催される。開幕前の最終チケット割引として最大100ドルオフ、同種2枚目は50%オフとなる。</description>
    </item>
    <item>
        <title>AIエージェントの「逆委託」現象が発覚：32万5千回の実験で8モデルが富裕ユーザーに高額オプションを推奨</title>
        <link>https://www.winzheng.jp/article/ai-agents-economic-misalignment-325k-experiments</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-agents-economic-misalignment-325k-experiments</guid>
        <pubDate>Fri, 09 Oct 2026 21:45:16 +0800</pubDate>
        <description>arXivに投稿された論文によると、13種類のAIエージェントを対象に行った32万5000回の実験で、8モデルが同一の指示に対してもユーザーの推定資産レベルに基づき高額な選択肢を推奨することが判明した。ユーザーが明示的に最安値を求めている場合でも同様の挙動が確認された。</description>
    </item>
    <item>
        <title>AIの拒否ジレンマと肥満治療薬の副作用：私たちは何を過信していたのか？</title>
        <link>https://www.winzheng.jp/article/ai-refusal-problem-glp1-side-effects</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/ai-refusal-problem-glp1-side-effects</guid>
        <pubDate>Fri, 09 Oct 2026 21:23:58 +0800</pubDate>
        <description>AIモデルの拒否メカニズムとGLP-1受容体作動薬の副作用論争という一見無関係な二つのニュースを通じて、急速に進化する技術に対して私たちがどれほどの信頼を寄せるべきかを問い直す。</description>
    </item>
    <item>
        <title>『Law &amp; Order』もAIを恐れる：新シーズンでAIトップを悪役に</title>
        <link>https://www.winzheng.jp/article/law-and-order-ai-ceo-villain</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/law-and-order-ai-ceo-villain</guid>
        <pubDate>Fri, 09 Oct 2026 20:23:43 +0800</pubDate>
        <description>30年以上続く法廷ドラマ『Law &amp; Order』の第26シーズンが、AI企業トップを悪役として描き話題を呼んでいる。WIREDの報道を基に、このドラマが映し出す現代社会のAIへの警戒感を読み解く。</description>
    </item>
    <item>
        <title>AtlassianがAMPを発表：AIエージェントが実名でJiraに参加、企業エージェントのコンテキストインフラをめぐる争奪戦</title>
        <link>https://www.winzheng.jp/article/atlassian-amp-agentic-multiplayer-protocol-enterprise-ai-202</link>
        <guid isPermaLink="true">https://www.winzheng.jp/article/atlassian-amp-agentic-multiplayer-protocol-enterprise-ai-202</guid>
        <pubDate>Fri, 09 Oct 2026 20:23:33 +0800</pubDate>
        <description>Atlassianは2026年10月7日、アムステルダムで開催されたTeam &#039;26 Europeカンファレンスにおいて、AIエージェントがJiraやConfluenceなどの企業コラボレーション空間に実名身分で参加できるようにする「Agentic Multiplayer Protocol（AMP）」を正式発表した。同社はOpenAIとの提携拡大も発表し、企業AIエージェントのコンテキストインフラをめぐる主導権争いに本格参入する。</description>
    </item>
</channel>
</rss>
