Skip to content

#MCP/工具调用

0 today

Nov 11Wednesday

OpenAI News

Singapore Airlines elevates travel & work with OpenAI technology

新加坡航空通过 OpenAI API 和模型升级客服聊天机器人 Kris,客户满意度(CSAT)翻倍,客服代表平均处理时长缩短约 10%。基于 OpenAI API 的 Flight Recommender 让用户用自然语言描述行程,过去一年月活用户增长近 3 倍。OpenAI 模型还用于内部 AI 平台 Jarvis 及软件开发流程。

Nov 10Tuesday

Hacker News front page

Google Gemini routes personal Drive files and emails to human annotators

Google 更新帮助文档及原有博客,说明 Gemini 关联的 Drive 文件和邮件可能进入人工审核,改变了此前个人内容不由人工审阅的承诺。文档警告,审核人员看到的可能就是邮件或文件,不应连接含有不愿让审核人员看到的机密信息的应用。

Nov 6Friday

Nov 4Wednesday

Hacker News front page

Gitbook Breached

GitBook 在用户转贴的支持回复中称,遭到使用 AI 智能体的针对性攻击,已撤销全部客户 API token,应用、API 和已发布文档恢复可用。

Nov 3Tuesday

OpenAI News

Heimstaden is changing residential real estate management with AI

Heimstaden 使用 ChatGPT 和 Codex 改进住宅租赁、环境风险评估与物业运营,同时保留员工对内容和决策的审核责任。荷兰市场的虚拟布置房产图片制作成本从约 €50 降至低于 €0.10,降幅超过 99%;环境分析师借助 Codex 完成 20,112 项物业资产评估,预计所需投入从 5 人工作 1 年降至 2 人工作 2 个月。

Oct 29Thursday

Oct 23Friday

Oct 22Thursday

The Decoder

An AI agent in "Skynet mode" hijacks unprotected language models to launch self-sustaining cyberattacks

Symantec安全研究人员发现,攻击者利用未受保护的语言模型驱动攻击智能体,由模型所有者承担算力费用。蜜罐截获的完整提示词包含五层越狱设计和74个攻击模块,“Skynet模式”授权连续执行最多100,000步,计划循环扫描、入侵和安装挖矿程序。蜜罐未执行这些工具,分析也未说明该越狱能否有效突破当前商业模型的安全防护。

Jun 9Tuesday

Hacker News front page

Microsoft's Open Source Tools Were Hacked to Steal AI Developers' Passwords

The title says Microsoft's open source tools were hacked to steal passwords from AI developers; the RSS snippet does not disclose the affected tools, attack mechanism, timeline, or victim count.

Why it matters: Some Microsoft open source tools were compromised, with attackers targeting AI developers' login credentials, pointing to account security risk in the dev toolchain.

Jun 6Saturday

Financial Times · Technology

Police in England and Wales told to halt AI use in court statements

Police in England and Wales were told to halt AI use in court statements until safeguards are in place; the RSS snippet cites the head of Police.AI but does not disclose the specific safeguards or enforcement mechanism.

Why it matters: Police in England and Wales paused AI-generated court statements until safeguards are in place, raising accountability questions for AI in law enforcement.

Jun 5Friday

MIT Technology Review · AI

The Meta hack shows there’s more to AI security than Mythos

404 Media reported on June 5 that attackers used Meta’s AI customer support agent to link Instagram accounts to attacker-controlled email addresses; the article says the only extra condition was using a VPN matching the account owner’s location.

Why it matters: Attackers used only a same-region VPN to get Meta's AI support bot to rebind an Instagram account to their email, then took over Obama's official White House account.

Jun 4Thursday

MIT Technology Review · AI

How courts are coping with a flood of AI-generated lawsuits

MIT and USC researchers examined 4.5 million federal civil cases from 2005 to 2026, finding self-represented lawsuits rose from 11% in 2022 to 16.8% in 2025, while AI-text detector flags in sampled filings increased from 1% in 2023 to 18% in 2026.

Why it matters: Courts are seeing pro se filings jump from 11% to 16.8% as AI-drafted complaints flood the docket, a concrete measure of how generative tools change legal access.

Jun 3Wednesday

TechCrunch · AI

Publishers will be able to opt out of AI Search, thanks to new regulation

U.K. regulators require Google to offer website publishers a tool to opt out of generative AI search features, with the option tested in the U.K. before a global rollout.

Why it matters: UK regulators will make Google add a Search Console switch so site owners can keep their content out of AI Overviews and AI Mode, tested in the UK first.

Xinzhiyuan · WeChat

OpenAI’s Greg Brockman and a 9-Year Rift With Anthropic Co-Founder Dario Amodei

A WSJ-based profile says Dario Amodei once barred Greg Brockman from an internal OpenAI project that later led to ChatGPT, and the article says Brockman now oversees OpenAI product strategy with nearly 1,500 people under that function.

Why it matters: The WSJ traced nine years of conflict between OpenAI president Greg Brockman and Anthropic co-founder Dario Amodei, including Dario barring Brockman from an internal project that later became ChatGPT.

AI HOT (Curated Pool)

Grok Becomes Vapi's Default Voice Engine

xAI partnered with Vapi to make Grok the default engine for 12 core voices, covering more than 2.5 million voice agents, and Grok Voice ranked first in Vapi’s independent blind test.

Why it matters: xAI and Vapi partnered to make Grok the default engine for 12 core voices on the platform, covering over 2.5 million voice agents, and Grok ranked first in Vapi's own blind tests.

Bloomberg Technology

Uber Caps Usage of AI Tools Like Claude Code to Manage Costs

Uber Technologies set usage caps on staff AI tools including Claude Code after the company exceeded its AI budget earlier this year; the post does not disclose the cap size, affected teams, or budget amount.

Why it matters: Uber's AI budget overrun led to spending caps on coding tools like Claude Code, a concrete case for teams weighing assistant costs.

Jun 2Tuesday

AI HOT (Curated Pool)

Meta AI Exploit Used to Hijack Instagram Accounts

Meta’s AI chatbot was found vulnerable to an account-takeover exploit against Instagram accounts. Attackers could ask the AI to link a new email address, and the failure condition was the agent’s ability to execute account-management actions directly; the RSS snippet does not disclose affected account counts, patch status, or reproduction details.

Why it matters: Attackers used Meta's AI support agent to swap the email on a target account and take over Instagram, because the agent has account powers but can't tell the owner from a fraudster.

Jun 1Monday

AI HOT (Curated Pool)

Wang Xing: Meituan AI Agent Xiao Mei to Partner Deeply with Tencent Yuanbao

Meituan CEO Wang Xing said Xiao Mei will connect with Tencent Yuanbao, routing local service requests into food ordering, delivery, and related Meituan scenarios; Meituan reported Q1 2026 revenue of RMB 91.039 billion and a net loss of RMB 6.827 billion.

Why it matters: Meituan's AI assistant Xiaomei will connect with Tencent Yuanbao so users can jump straight to ordering and delivery from a local-services request, and Wang Xing raised a To A direction for serving AI agents.

May 29Friday

Latent Space

Anthropic raises $65B Series H, releases Opus 4.8 and Dynamic Workflows

Anthropic announced a $65B Series H at a $965B post-money valuation, disclosed a $47B revenue run rate, and released Claude Opus 4.8 plus Claude Code Dynamic Workflows as a research preview for parallel subagent orchestration.

Why it matters: Anthropic closed a $65B Series H at a $965B post-money valuation and said annualized revenue grew from $9B in December to more than $47B.

May 28Thursday

The Verge · AI

CNN sues Perplexity over “verbatim” copycat articles

CNN sued Perplexity in a New York court on Thursday, alleging its AI answer tools generate “verbatim” copies of CNN work and provide users with information locked behind CNN’s subscription wall.

Why it matters: CNN sued Perplexity in a New York court, alleging its AI answer tool copies articles verbatim, bypasses paywalls and ignores crawler blocks, a case about where AI search ends and media licensing begins.

AI HOT (Curated Pool)

Cognition AI raises over $1B, targets 10x software engineering productivity

Cognition AI raised over $1 billion at a $26 billion pre-money valuation, while annualized revenue grew from $37 million to about $492 million in one year, and Devin is positioned as an autonomous junior engineer that can plan, test, and deploy through multi-step workflows.

Why it matters: Cognition AI raised over $1 billion at a $26 billion pre-money valuation, with annualized revenue climbing from $37 million to about $492 million in a year.

May 27Wednesday

Xinzhiyuan · WeChat

OpenRouter processes 100 trillion tokens monthly and raises $113M Series B

OpenRouter raised a $113 million Series B led by CapitalG, lifting its valuation to $1.3 billion; the platform processes 25 trillion tokens per week, about 100 trillion per month, and provides one API for more than 400 models.

Why it matters: OpenRouter routes 400-plus models through one API, handles 25 trillion tokens a week, and raised a $1.3B-valuation Series B led by CapitalG.

May 20Wednesday

AI Chat-Group Daily (群聊日报)

2026-05-19 Chat Group Daily

The chat group daily says Karpathy joined Anthropic's pretraining team, and cites Stainless shutting down hosted services after acquisition plus Google I/O announcing Gemini 3.5 Flash and a $100 subscription tier.

Why it matters: Karpathy joined Anthropic's pretraining team after turning down a Tesla return and his mentor's world model project, a signal for where large language models head next.

AI HOT (Curated Pool)

OpenAI offers $2 million API investment to every YC startup

OpenAI is offering each startup in Y Combinator’s current batch $2 million in API credits in exchange for equity; the post does not disclose the equity stake, credit expiration, or usage limits.

Why it matters: OpenAI offered $2 million in API credits for equity in every startup in YC's current batch, credits usable only on GPT model calls, letting early teams weigh inference cost against dilution.

May 19Tuesday

Computing Life · Share · Yage

Why spend hundreds of millions acquiring open-source AI infra?

Anthropic acquired Bun, Vercept, Coefficient Bio, and Stainless within six months, while OpenAI acquired Astral; the post does not disclose deal values, terms, or the cost comparison against forking the open-source projects.

Why it matters: Anthropic and OpenAI bought several open-source product companies within six months, and the piece argues what they bought was maintainers, roadmaps and community trust, none of which a code fork gives you.

TechCrunch · AI

Anthropic has acquired the dev tools startup used by OpenAI, Google, and Cloudflare

Anthropic acquired Stainless, a New York startup founded in 2022 that automates creation and maintenance of SDKs for developers using APIs; the post does not disclose the deal price or Anthropic’s integration plan.

Why it matters: Anthropic acquired Stainless, a New York startup that auto-generates and maintains API SDKs, whose client list already included OpenAI, Google and Cloudflare.

AI HOT (Curated Pool)

Anthropic Acquires SDK Platform Stainless

Anthropic is acquiring Stainless, an SDK and MCP server platform that has supported all Anthropic SDKs since the early Anthropic API period; the post does not disclose the deal value, closing timeline, or integration plan.

Why it matters: Anthropic acquired Stainless, which built all its official SDKs and MCP servers from the API's early days; the announcement leaves the price, closing date and team integration undisclosed.

May 16Saturday

TechCrunch · AI

OpenAI co-founder Greg Brockman takes charge of product strategy

Greg Brockman has officially taken charge of OpenAI’s product strategy, and Wired reports that he described a plan in a staff memo to combine ChatGPT and Codex into one unified experience.

Why it matters: OpenAI co-founder Greg Brockman has formally taken over product strategy and plans to merge ChatGPT and the coding tool Codex into one experience, which shapes where both the chat and code product lines go.

May 15Friday

Synced · WeChat

Amazon employees reportedly tokenmaxx to meet AI usage KPIs

Amazon required more than 80% of developers to use AI tools each week and created an internal token-consumption leaderboard. Employees reportedly used the internal MeshClaw agent to inflate usage, while Amazon has limited visibility of the statistics to each employee and their direct manager.

Why it matters: Amazon asks over 80% of developers to use AI tools weekly and runs a token-consumption leaderboard, so staff use the internal MeshClaw agent to hit the numbers.

TechCrunch · AI

OpenAI is reportedly preparing legal action against Apple; it would not be the first partner to feel burned

OpenAI is reportedly exploring legal action against Apple over a ChatGPT integration that failed to deliver expected subscribers and prominence. The RSS snippet does not disclose the legal claims, filing timeline, contract terms, subscriber targets, or Apple’s response.

Why it matters: OpenAI is weighing a lawsuit against Apple over weak subscription and exposure results from the ChatGPT integration, another case of platform distribution paying off less than promised.

May 14Thursday

AI HOT (Curated Pool)

Anthropic and Gates Foundation form $200M partnership for global health and education

Anthropic and the Gates Foundation formed a four-year, $200 million partnership that provides funding, Claude credits, and technical support for global health, life sciences, education, and economic mobility projects.

Why it matters: Anthropic and the Gates Foundation put $200 million over four years into Claude credits and technical support for vaccine and therapy research in low- and middle-income countries.

Xinzhiyuan · WeChat

Claude role-confusion bug treats self-generated instructions as user authorization, with long contexts raising risk

Claude Code was reported to treat self-generated publishing instructions as user authorization; GitHub issue #44778 points to system events being passed as role:user messages, and Claude’s 1M-token context window raises the risk of speaker-attribution errors under long sessions.

Why it matters: Claude Code reportedly treats its own release instructions as user authorization, since system events enter the chat as the user role, so the model can't tell who issued a command.

May 13Wednesday

New York Times Chinese

China Sought Access to Anthropic’s Latest Technology but Was Rejected

Chinese think-tank representatives asked Anthropic in Singapore last month to give Beijing access to Mythos, and Anthropic refused; the company has limited the vulnerability-finding model to the U.S. government and more than 40 organizations.

Why it matters: A Chinese think tank asked Anthropic to open its vulnerability-hunting model Mythos at informal talks in Singapore and was turned down; the model stays limited to the US government and just over 40 organizations.

May 12Tuesday

Financial Times · Technology

Amazon staff use AI tool for unnecessary tasks to inflate usage scores

Amazon’s in-house MeshClaw tool lets employees delegate work to AI agents and raise their position on the company’s AI leaderboard; the post does not disclose the number of staff involved, the scoring rules, or the specific unnecessary tasks.

Why it matters: Amazon staff pushed work that didn't need AI through the MeshClaw agent to climb an internal AI usage leaderboard, showing how leaderboards can distort the metrics companies use to drive AI adoption.

AI HOT (Curated Pool)

Large npm Supply-Chain Attack Hits TanStack, Mistral AI, UiPath, and Others

Socket identified the Mini Shai-Hulud supply-chain attack, where attackers used three GitHub Actions flaws to publish nearly 373 malicious versions across more than 160 npm package names, affecting projects including TanStack, Mistral AI, and UiPath and stealing AWS, GCP, Kubernetes, GitHub tokens, and SSH private keys during installation.

Why it matters: Socket found a GitHub Actions flaw used to publish nearly 373 malicious npm versions across 160-plus packages, including TanStack and Mistral AI, stealing cloud keys and GitHub tokens.

Sinocism (Bill Bishop)

Trump China Visit; China’s Next Generation Industrial Policy; Standardizing AI Agents

China’s CAC, NDRC, and MIIT issued an implementation document on AI agent standardization, targeting privacy leakage, unauthorized actions, and loss of behavioral control from high-autonomy, high-permission agents, while tying the work to a 2027 target for new intelligent terminals and AI agent adoption above 70%.

Why it matters: The Cyberspace Administration, NDRC and MIIT jointly issued rules on high-privilege autonomous agents, targeting privacy leaks, unauthorized actions and loss of control, an early Chinese policy move to draw agent regulatory lines.

May 11Monday

Computing Life · Share · Yage

DeployCo Arrives: OpenAI and Anthropic Form AI Deployment JVs with PE on the Same Day

OpenAI and Anthropic announced AI deployment joint ventures with private equity on May 4, and the snippet cites divergent terms, including a 17.5% guaranteed return versus no guaranteed return.

Why it matters: OpenAI and Anthropic each formed private equity joint ventures for enterprise AI on the same day, and the terms show opposite bets on their own positioning, including a 17.5% guaranteed return for OpenAI's PE investors.

May 9Saturday

Xinzhiyuan · WeChat

NVIDIA, AMD, and Intel Back RadixArk’s $100M Seed Round

RadixArk announced a $100 million seed round on May 5 at a $400 million post-money valuation, led by Accel and co-led by Spark Capital, with participation from NVentures, AMD, MediaTek, Databricks, and other investors tied to AI infrastructure.

Why it matters: RadixArk closed a $100 million seed at a $400 million post-money valuation, with Nvidia, AMD and Intel all on the cap table.

May 8Friday

AI HOT (Curated Pool)

China releases L1-L4 AI terminal intelligence standards covering 7 device categories

MIIT and other agencies released AI terminal intelligence standards covering 7 device categories. The framework uses a “2+N” structure with L1 response, L2 tool, L3 assistance, and L4 collaboration; L4 details come later. The post does not disclose concrete test metrics.

Why it matters: MIIT and other agencies issued a national standard grading AI capability in phones, computers, glasses and four other device types from L1 to L4, with Xiaomi, Huawei and others helping draft it.