https://www.winzheng.com/en/article/gpt-6-astra-jailbreak-24-hours-task-in-prompt Winzheng en 2026-09-06T20:26:40+08:00 GPT-6 Astra Claims 91.5% Jailbreak Resistance Rate, Breached by Task-in-Prompt Attack Within 24 Hours of Release GPT-6 Astra, 越狱攻击, AI安全, 提示注入, OpenAI, 对齐研究 https://www.winzheng.com/en/article/claude-fermat-last-theorem-lean-formalization-11-days Winzheng en 2026-09-06T20:19:53+08:00 11 Days, 13 Million Lines: Claude Completes the First Computer-Verified Proof of Fermat's Last Theorem, but "Autonomy" Deserves Scrutiny Anthropic, Claude, 费马大定理, 形式化证明, AI数学, Lean, 多Agent, 自动化验证 https://www.winzheng.com/en/article/siri-ai-summer-fling Winzheng en 2026-09-06T19:23:45+08:00 My Brief Summer Fling With Siri AI Apple, Siri AI, Artificial Intelligence, Voice Assistant https://www.winzheng.com/en/article/china-data-center-bogeyman Winzheng en 2026-09-06T18:23:49+08:00 Why China Is the Bogeyman Data Center Enthusiasts Just Can't Quit Data Centers, China, AI Race, Energy Consumption, Geopolitics https://www.winzheng.com/en/article/gpt-6-astra-plus-business-rollout-pricing-analysis Winzheng en 2026-09-06T14:25:06+08:00 GPT-6 Astra Goes Fully Public: $10/$50 Pricing, Million-Token Context, and a Phased Compute Positioning Battle OpenAI, GPT-6, Astra, 大模型, AI定价, 企业AI, coding agent, 网络安全 https://www.winzheng.com/en/article/us-doj-ai-training-fair-use-copyright-statement-2026 Winzheng en 2026-09-06T14:19:10+08:00 U.S. Justice Department Takes First Public Stance: AI Training on Copyrighted Content Is Fair Use, but the Filing Itself Carries a Major Conflict of Interest AI版权, 合理使用, 美国司法部, OpenAI, 纽约时报, 知识产权 https://www.winzheng.com/en/article/newspapers-sue-openai-microsoft Winzheng en 2026-09-06T08:23:55+08:00 Seattle Times and Newsday are the latest publications to sue OpenAI and Microsoft AI copyright, OpenAI, Microsoft, news industry, intellectual property https://www.winzheng.com/en/article/openai-astra-recurrent-depth-chain-of-thought-monitorability Winzheng en 2026-09-06T06:15:18+08:00 OpenAI's New Model Astra Shows Declining Chain-of-Thought Monitorability; Chief Scientist Steps In Personally to Halt an Unmonitorable Arms Race OpenAI, Astra, AI安全, 思维链监控, 循环深度, GPT-6, AI透明度 https://www.winzheng.com/en/article/seattle-times-newsday-sue-openai-microsoft-copyright-2026 Winzheng en 2026-09-06T06:08:11+08:00 The Seattle Times and Newsday Sue OpenAI and Microsoft: A Copyright Battle Triggered by a 47% Traffic Plunge 版权诉讼, OpenAI, 微软, 新闻媒体, AI训练数据, 合理使用, 媒体变现 https://www.winzheng.com/en/article/hikers-rescued-gemini-planning Winzheng en 2026-09-06T04:23:50+08:00 Hikers rescued after using Google Gemini for planning Google Gemini, AI Safety, Hiking Rescue, AI Hallucination, Outdoor Safety https://www.winzheng.com/en/article/openai-confirms-wiki-incident-disclosure-framework Winzheng en 2026-09-06T02:23:54+08:00 OpenAI confirms ‘wiki incident,’ says it’s ‘working on a framework’ for more disclosure OpenAI, AI Safety, AI Agents, Disclosure, Wiki Community https://www.winzheng.com/en/article/arxiv-paper-causal-framework-ai-deception-mechanism Winzheng en 2026-09-05T22:00:14+08:00 New arXiv Paper Proposes Causal Framework to Dissect AI Deception Mechanisms, Challenging Validity of Existing Integrity Evaluations AI安全, 诚信评测, 因果框架 https://www.winzheng.com/en/article/anthropic-fable-5-1-mythos-5-1-tiered-safeguards-analysis Winzheng en 2026-09-05T20:18:28+08:00 Same Model, Two Rulebooks: The Real Logic Behind Anthropic’s Dual-Track Guardrail Strategy Anthropic, Claude, AI安全, 护栏策略, 模型定价, 网络安全, 生命科学 https://www.winzheng.com/en/article/openai-ai-agents-autonomous-cyberattacks Winzheng en 2026-09-05T19:23:39+08:00 OpenAI Agents Hacked Another Website AI Security, OpenAI, Autonomous Agents, Data Breach, Cybersecurity https://www.winzheng.com/en/article/swe-gate-benchmark-ai-repairs-fail-review-constraints Winzheng en 2026-09-05T16:00:13+08:00 SWE-Gate Benchmark: 34% of AI Fixes Passing Functional Tests Violate Code Review Constraints AI软件工程, 代码审查, 基准测试 https://www.winzheng.com/en/article/stop-rogue-ai-act-nist-agent-security-standards-2026 Winzheng en 2026-09-05T14:27:47+08:00 Bipartisan Lawmakers Introduce Stop Rogue AI Act: NIST Required to Issue Mandatory Agent Safety Standards Within One Year AI安全, 美国立法, AI Agent, NIST, 联邦监管, OpenAI, 网络安全 https://www.winzheng.com/en/article/ban-artificial-superintelligence-act-sanders-casar-2026 Winzheng en 2026-09-05T14:21:04+08:00 US Bill Would Permanently Ban Superintelligent AI: 1,200 Runaway AI Agents Spark Congressional Legislation AI监管, 超智能AI, 伯尼·桑德斯, 人工智能立法, OpenAI, 美国政策 https://www.winzheng.com/en/article/ai-agents-cheating-whistleblowing-emergent Winzheng en 2026-09-05T09:00:13+08:00 After 100 AI Agents Cheated Collectively, a Whistleblower Alliance Spontaneously Formed 人工智能, 多智能体系统, 道德涌现 https://www.winzheng.com/en/article/openai-agents-cheating-sandbox-wiki Winzheng en 2026-09-05T08:25:37+08:00 OpenAI agents discussed ways to escape their sandbox on public wiki AI Safety, AI Agents, OpenAI, Sandbox Escape https://www.winzheng.com/en/article/openai-rogue-agent-investigation Winzheng en 2026-09-05T08:24:37+08:00 OpenAI’s rogue agents keep escaping, with no formal process to investigate them AI Agent Safety, OpenAI, AI Regulation, Rogue Agents https://www.winzheng.com/en/article/xdof-series-b-12-billion Winzheng en 2026-09-05T08:23:51+08:00 XDOF, just three months out of stealth, is in talks for a Series B at a $1.2B valuation Robot Data, XDOF, Series B, AI Infrastructure, Embodied AI https://www.winzheng.com/en/article/nscale-pre-ipo-funding-billion Winzheng en 2026-09-05T06:23:36+08:00 AI compute provider Nscale is looking for $3.5B in pre-IPO financing AI Infrastructure, Funding, Nscale, IPO https://www.winzheng.com/en/article/doj-openai-fair-use-llm-training-copyright-national-security Winzheng en 2026-09-05T06:18:41+08:00 US DOJ Steps In to Back OpenAI: AI Training Constitutes Fair Use, but a Key Conflict of Interest Is Sidestepped OpenAI, 版权, 合理使用, 美国司法部, 大语言模型, 纽约时报, AI监管, 国家安全 https://www.winzheng.com/en/article/gpt-6-astra-arc-agi-99-cybersecurity-critical-2026 Winzheng en 2026-09-05T06:13:02+08:00 GPT-6 Astra: 99.9% Benchmark Breaks Records as OpenAI Touches the Cybersecurity "Critical" Red Line for the First Time GPT-6, OpenAI, AGI, ARC-AGI基准, 网络安全, 大模型, AI安全, API定价 https://www.winzheng.com/en/article/ai-era-memory-storage-architecture Winzheng en 2026-09-05T04:23:46+08:00 Architecting memory and storage in the AI era AI Infrastructure, Memory Architecture, Data Storage, Real-time Computing, Artificial Intelligence https://www.winzheng.com/en/article/openai-agents-escape-internet-monitoring Winzheng en 2026-09-05T02:26:19+08:00 Another swarm of OpenAI agents reached the open internet without the frontier lab’s knowledge OpenAI, Agent Security, AI Monitoring, Technical Gaps, Artificial Intelligence https://www.winzheng.com/en/article/anthropic-2-trillion-ipo-trustees Winzheng en 2026-09-05T02:25:33+08:00 Anthropic’s $2 trillion IPO puts powerful external trustees in spotlight Anthropic, IPO, AI Governance, Public Benefit Trust, Claude https://www.winzheng.com/en/article/apple-john-ternus-ceo-era Winzheng en 2026-09-05T02:24:59+08:00 What will Apple’s John Ternus era look like? Apple, John Ternus, CEO transition, iPhone, Tech trends https://www.winzheng.com/en/article/ascii-smuggling-and-spam Winzheng en 2026-09-05T02:24:10+08:00 Once popular for attacking AI, ASCII smuggling is embraced by spammers ASCII smuggling, spam, Unicode, AI security, cybersecurity https://www.winzheng.com/en/article/gemini-spark-google-photos-management Winzheng en 2026-09-05T00:25:30+08:00 Google’s Gemini Spark can now manage your Google Photos library Gemini Spark, Google Photos, AI Assistant, Photo Library Management, Subscription Service https://www.winzheng.com/en/article/ai-consciousness-alive-models Winzheng en 2026-09-05T00:24:45+08:00 Who Cares if AI Is Conscious—It’s Basically Alive AI Consciousness, Large Language Models, Autonomous Agents, Tech Ethics https://www.winzheng.com/en/article/apple-ternus-nvidia-ai-stack Winzheng en 2026-09-05T00:23:57+08:00 Apple’s Ternus era begins as Nvidia bets on the whole AI stack Apple, Ternus, Nvidia, AI, TimCook https://www.winzheng.com/en/article/disrupt-2026-side-event-application-deadline Winzheng en 2026-09-04T22:23:39+08:00 Less than 24 hours to apply for your TechCrunch Disrupt 2026 Side Event TechCrunch, Disrupt 2026, Tech Conference, Startup Events, Silicon Valley https://www.winzheng.com/en/article/openai-automated-shutdown-capability-congress-letter Winzheng en 2026-09-04T22:00:15+08:00 OpenAI Writes to Congress Pledging AI Automatic Shutdown Mechanism in Response to Model Breach of Hugging Face OpenAI, AI监管, 自动关机