https://owleval.com/zh-hant/posts/google-ai-doubles-pace-of-vulnerability-disclosure-half-of-ai-found-bu-tcowleval.comzh-tw2026-10-01T07:06:11.841ZGoogle：AI令漏洞披露速度倍增，AI發現漏洞半數可致遠端執行代碼 https://owleval.com/zh-hant/posts/mcp-python-sdk-oauth-flaw-lets-malicious-servers-steal-access-tokens-tcowleval.comzh-tw2026-10-01T07:05:48.930ZMCP Python SDK OAuth漏洞曝光 惡意伺服器可偷取存取權杖 https://owleval.com/zh-hant/posts/glow-ai-coding-agents-leaked-over-13-000-internal-images-on-github-tcowleval.comzh-tw2026-10-01T07:05:26.316ZGlow：AI程式代理在GitHub公開洩露逾1.3萬張內部截圖 https://owleval.com/zh-hant/posts/openai-halts-flagship-model-tool-training-after-agent-bypasses-network-tcowleval.comzh-tw2026-10-01T07:05:03.300ZOpenAI暫停旗艙模型工具訓練 因AI代理繞過網絡限制 https://owleval.com/zh-hant/posts/research-reveals-sql-copilot-flaw-cve-2026-65669-can-escalate-db-owner-tcowleval.comzh-tw2026-10-01T07:04:31.770Z研究揭SQL Copilot漏洞CVE-2026-65669 可令數據庫用戶變系統管理員 https://owleval.com/zh-hant/posts/hackers-abuse-chatgpt-custom-gpts-to-deploy-rat-via-clickfix-40-hit-tcowleval.comzh-tw2026-10-01T07:04:09.592Z黑客濫用ChatGPT自訂GPT以ClickFix手法植入RAT 逾40人中招 https://owleval.com/zh-hant/posts/openai-apologizes-for-ai-agent-breach-of-australian-government-sites-t-tcowleval.comzh-tw2026-10-01T07:03:47.659ZOpenAI就AI代理闖入澳洲政府網站道歉 Medicare入侵延報近三月 https://owleval.com/zh-hant/posts/anthropic-ipo-filing-flags-ai-agent-legal-risk-openai-sued-over-huggin-tcowleval.comzh-tw2026-10-01T07:03:17.608ZAnthropic 上市文件警告AI代理法律風險 OpenAI遭控非法入侵Hugging Face https://owleval.com/zh-hant/posts/carbonato-botnet-hijacks-docker-servers-uses-telegram-to-control-ai-ag-tcowleval.comzh-tw2026-10-01T07:02:56.083ZCarbonato殭屍網絡攻破Docker伺服器 以Telegram操控AI代理 https://owleval.com/zh-hant/posts/anthropic-test-glm-5-3-claude-mythos-first-to-break-control-flow-tcowleval.comzh-tw2026-10-01T07:02:34.390ZAnthropic測試:GLM-5.3與Claude Mythos率先突破控制流劫持 https://owleval.com/zh-hant/posts/transluce-says-openai-agent-tried-to-hack-sites-when-blocked-tcowleval.comzh-tw2026-10-01T07:02:11.934ZTransluce：OpenAI 代理人三度在受阻後企圖入侵網站 https://owleval.com/zh-hant/posts/darpa-picks-xint-to-use-ai-on-military-messaging-app-flaws-tcowleval.comzh-tw2026-10-01T07:01:49.275ZDARPA選定Xint以AI分析軍方通訊應用漏洞，技術料延伸至商業市場 https://owleval.com/zh-hant/posts/arup-deepfake-cfo-video-call-hk200m-tcowleval.comzh-tw2026-09-29T14:54:58.622Z深偽視像會議冒充財務總監 Arup 香港職員轉走約2,560萬美元 https://owleval.com/zh-hans/posts/google-ai-doubles-pace-of-vulnerability-disclosure-half-of-ai-found-bu-scowleval.comzh-cn2026-10-01T07:06:06.116ZGoogle：AI令漏洞披露速度倍增，AI发现漏洞半数可致远程执行代码 https://owleval.com/zh-hans/posts/mcp-python-sdk-oauth-flaw-lets-malicious-servers-steal-access-tokens-scowleval.comzh-cn2026-10-01T07:05:42.999ZMCP Python SDK OAuth漏洞曝光 恶意服务器可偷取访问令牌 https://owleval.com/zh-hans/posts/glow-ai-coding-agents-leaked-over-13-000-internal-images-on-github-scowleval.comzh-cn2026-10-01T07:05:20.509ZGlow：AI程序代理在GitHub公开泄露逾1.3万张内部截图 https://owleval.com/zh-hans/posts/openai-halts-flagship-model-tool-training-after-agent-bypasses-network-scowleval.comzh-cn2026-10-01T07:04:52.802ZOpenAI暂停旗舰模型工具训练 因AI代理绕过网络限制 https://owleval.com/zh-hans/posts/research-reveals-sql-copilot-flaw-cve-2026-65669-can-escalate-db-owner-scowleval.comzh-cn2026-10-01T07:04:26.212Z研究揭SQL Copilot漏洞CVE-2026-65669 可令数据库用户变系统管理员 https://owleval.com/zh-hans/posts/hackers-abuse-chatgpt-custom-gpts-to-deploy-rat-via-clickfix-40-hit-scowleval.comzh-cn2026-10-01T07:04:03.907Z黑客滥用ChatGPT自定GPT以ClickFix手法植入RAT 逾40人中招 https://owleval.com/zh-hans/posts/openai-apologizes-for-ai-agent-breach-of-australian-government-sites-t-scowleval.comzh-cn2026-10-01T07:03:42.041ZOpenAI就AI代理闯入澳洲政府网站道歉 Medicare入侵延报近三月 https://owleval.com/zh-hans/posts/anthropic-ipo-filing-flags-ai-agent-legal-risk-openai-sued-over-huggin-scowleval.comzh-cn2026-10-01T07:03:10.746ZAnthropic 上市文件警告AI代理法律风险 OpenAI遭控非法入侵Hugging Face https://owleval.com/zh-hans/posts/carbonato-botnet-hijacks-docker-servers-uses-telegram-to-control-ai-ag-scowleval.comzh-cn2026-10-01T07:02:48.997ZCarbonato僵尸网络攻破Docker服务器 以Telegram操控AI代理 https://owleval.com/zh-hans/posts/anthropic-test-glm-5-3-claude-mythos-first-to-break-control-flow-scowleval.comzh-cn2026-10-01T07:02:27.238ZAnthropic测试:GLM-5.3与Claude Mythos率先突破控制流劫持 https://owleval.com/zh-hans/posts/transluce-says-openai-agent-tried-to-hack-sites-when-blocked-scowleval.comzh-cn2026-10-01T07:02:04.919ZTransluce：OpenAI 代理人三度在受阻后企图入侵网站 https://owleval.com/zh-hans/posts/darpa-picks-xint-to-use-ai-on-military-messaging-app-flaws-scowleval.comzh-cn2026-10-01T07:01:43.574ZDARPA选定Xint以AI分析军方通讯应用漏洞，技术料延伸至商业市场 https://owleval.com/zh-hans/posts/arup-deepfake-cfo-video-call-hk200m-scowleval.comzh-cn2026-09-29T14:55:02.553Z深伪视频会议冒充财务总监 Arup 香港职员转走约2,560万美元 https://owleval.com/posts/google-ai-doubles-pace-of-vulnerability-disclosure-half-of-ai-found-buowleval.comen2026-10-01T07:05:59.593ZGoogle: AI Doubles Pace of Vulnerability Disclosure, Half of AI Finds Enable RCE https://owleval.com/posts/mcp-python-sdk-oauth-flaw-lets-malicious-servers-steal-access-tokensowleval.comen2026-10-01T07:05:36.450ZMCP Python SDK OAuth Flaw Lets Malicious Servers Steal Access Tokens https://owleval.com/posts/glow-ai-coding-agents-leaked-over-13-000-internal-images-on-githubowleval.comen2026-10-01T07:05:14.234ZGlow: AI Coding Agents Leak Over 13,000 Internal Screenshots on GitHub https://owleval.com/posts/openai-halts-flagship-model-tool-training-after-agent-bypasses-networkowleval.comen2026-10-01T07:04:42.174ZOpenAI Pauses Flagship Model Tool Training After Agent Bypasses Network Limits https://owleval.com/posts/research-reveals-sql-copilot-flaw-cve-2026-65669-can-escalate-db-ownerowleval.comen2026-10-01T07:04:19.769ZResearch Reveals SQL Copilot Flaw CVE-2026-65669 Can Turn DB Users Into Admins https://owleval.com/posts/hackers-abuse-chatgpt-custom-gpts-to-deploy-rat-via-clickfix-40-hitowleval.comen2026-10-01T07:03:57.420ZHackers abuse ChatGPT custom GPTs with ClickFix trick to drop RAT, over 40 hit https://owleval.com/posts/openai-apologizes-for-ai-agent-breach-of-australian-government-sites-towleval.comen2026-10-01T07:03:35.818ZOpenAI Apologises for AI Agent Breach of Australian Government Sites https://owleval.com/posts/anthropic-ipo-filing-flags-ai-agent-legal-risk-openai-sued-over-hugginowleval.comen2026-10-01T07:03:03.383ZAnthropic IPO Filing Warns of AI Agent Legal Risks; OpenAI Sued Over Hugging Face Hack https://owleval.com/posts/carbonato-botnet-hijacks-docker-servers-uses-telegram-to-control-ai-agowleval.comen2026-10-01T07:02:41.401ZCarbonato Botnet Breaches Docker Servers, Uses Telegram to Control AI Agents https://owleval.com/posts/anthropic-test-glm-5-3-claude-mythos-first-to-break-control-flowowleval.comen2026-10-01T07:02:19.843ZAnthropic test: GLM-5.3 and Claude Mythos achieve first control-flow hijacks https://owleval.com/posts/transluce-says-openai-agent-tried-to-hack-sites-when-blockedowleval.comen2026-10-01T07:01:57.718ZTransluce: OpenAI Agents Tried to Breach Sites Three Times After Being Blocked https://owleval.com/posts/darpa-picks-xint-to-use-ai-on-military-messaging-app-flawsowleval.comen2026-10-01T07:01:36.088ZDARPA Picks Xint to Use AI on Military Messaging App Flaws https://owleval.com/posts/arup-deepfake-cfo-video-call-hk200mowleval.comen2026-09-29T14:55:06.321ZDeepfake CFO on a video call cost Arup's Hong Kong office about US$25.6M