How we contain Claude across products
Simon Willison's Weblog2d agoCommentaryAnthropic explains how process sandboxes, VMs, filesystem boundaries, and egress controls limit what Claude agents can access. Claude.ai uses gVisor; local Claude Code uses Seatbelt on macOS and Bubblewrap on Linux; Cowork runs in a full VM. Simon Willison highlights the documentation quality, notes a previously missed file-exfiltration path, and plans to revisit Anthropic's open-source srt tool.
Running Python ASGI Apps in the Browser via Pyodide + a Service Worker
Simon Willison's Weblog2d agoTutorialSimon Willison demonstrates an experiment for running Python ASGI apps entirely in the browser using Pyodide and a Service Worker. The approach addresses a Datasette Lite limitation: HTML returned through intercepted navigation did not execute script tags, breaking features and plugins. Claude Opus 4.8, used through Claude Code for web, helped explore the implementation. Basic ASGI and Datasette 1.0a31 demos are available.
I Am Retiring from Tech to Live Offline
Simon Willison's Weblog3d agoEthicsSimon Willison highlights Chad Whitacre’s decision to leave tech and Open Source, framed not as a forum threat but as concrete action. Whitacre describes wanting to become “AI Amish” or “Internet Amish,” moving toward an offline, analog life closer to 1980 than 1780. A previous post about using Claude Code with Opus 4.5 shows how agentic AI felt intoxicating and unsettling enough to push him away from technological accelerationism.
Anthropic Releases Claude Opus 4.8 With Integrity Upgrades and Dynamic Workflows
INSIDE 硬塞 AI4d agoReleaseAnthropic released Claude Opus 4.8 as a rapid iteration focused on stronger integrity and reliability for high-risk tasks. The company also previewed Dynamic Workflows, a feature designed to coordinate multiple agents on large-scale jobs such as code migration. The article mentions Mythos entering a countdown toward unblocking, but does not provide detailed availability or product specifics.
Anthropic Series H Valuation Reaches $965B, Backed by Memory Giants★ 78
INSIDE 硬塞 AI4d agoBusinessAnthropic completed a $65 billion Series H round, bringing its valuation to $965 billion and reportedly surpassing OpenAI. The round included strategic investments from memory makers Micron, Samsung, and SK Hynix. The news highlights how frontier AI companies are increasingly tied to hardware and memory supply chains, as investors continue backing foundational model competition.
Anthropic raises $965B Series H, releases Opus 4.8 and Dynamic Workflows/ultracode
Latent Space4d agoBusinessThe visible AINews item centers on Anthropic, claiming a $965B Series H alongside Opus 4.8 and Dynamic Workflows/ultracode releases. The available body text is extremely brief, offering only the editorial line “Total Anthropic victory!” It signals a major Anthropic narrative across capital, Claude models, and developer workflows, but provides no detailed specs, benchmarks, investor terms, or availability information.
Claude Opus 4.8: "a modest but tangible improvement"
Simon Willison's Weblog4d agoReleaseAnthropic shipped Claude Opus 4.8, and Simon Willison highlights the unusually restrained release language: a “modest but tangible improvement.” The model keeps most Opus 4.7 pricing and specs, while evaluations suggest it is more likely to flag uncertainty and less likely to ignore flaws in code it wrote. Developer-relevant changes include mid-conversation system messages and a lower prompt-cache minimum of 1,024 tokens.
llm-anthropic 0.25.1
Simon Willison's Weblog4d agoReleaseSimon Willison released llm-anthropic 0.25.1 with support for the new Claude Opus 4.8 model, exposed as claude-opus-4.8. The release adds a -o fast 1 option for Anthropic fast mode, limited to organizations that have the feature enabled. It also changes default max_tokens behavior so each model now defaults to its maximum output instead of 8,192.
markdown-svg-renderer
Simon Willison's Weblog4d agoNew ToolSimon Willison shared markdown-svg-renderer, a customized Markdown rendering tool with special handling for fenced SVG code blocks. It renders the SVG image and also provides a tab for switching back to the source code. Users can paste Markdown directly or load a CORS-enabled Markdown file or Gist by URL, with an example using LLM pelican logs for Opus 4.8.
Trump loses more control over AI regulation as Illinois passes landmark law★ 74
Ars Technica AI5d agoRegulationIllinois lawmakers passed a landmark AI accountability bill requiring major frontier AI developers to publish safety frameworks, assess catastrophic risks, report incidents, and undergo third-party audits. OpenAI and Anthropic supported the measure, while industry groups warned that state-level rules could impose subjective compliance duties without national standards. The bill signals that states are continuing to fill the federal AI regulation gap despite Trump’s efforts to limit fragmented state oversight.
Anthropic releases Opus 4.8 with new Dynamic Workflows tool
TechCrunch AI5d agoReleaseAnthropic has released a new Opus model, Opus 4.8, alongside a tool called Dynamic Workflows. The report says the tool is designed to coordinate swarms of subagents, pointing to a focus on multi-agent orchestration. The source does not provide benchmarks, pricing, API details, availability, or concrete use cases.
Claude’s new model is more ‘honest’ when it messes up
The Verge AI5d agoReleaseAnthropic is releasing Claude Opus 4.8 and highlighting the model’s “honesty” as a key improvement. The company says it trains its models to avoid unsupported claims, addressing a broader issue where AI systems sometimes jump to conclusions. Based on the provided excerpt, the update is positioned around reliability and uncertainty handling rather than a specific new tool or benchmark result.
RSI is the new AGI — and it’s just as hard to pin down
TechCrunch AI5d agoCommentaryTechCrunch reports that recursive self-improvement, or RSI, is becoming a new AI industry fixation, much like AGI. Researchers and startups including Recursive Superintelligence, Auto-Research, AutoScientist, and Disarray are exploring ways for AI systems to automate parts of AI research. But experts caution that AI-assisted research is not the same as fully autonomous self-improvement, especially while models still struggle with long-term self-direction and verification.
Choosing to Stay Human★ 74
One Useful Thing (Mollick)6d agoCommentaryEthan Mollick warns that frictionless AI use can produce hollow writing, weaken learning, and encourage cognitive surrender. He contrasts poor uses of ChatGPT that shortcut effort with tutor-like AI systems that improve learning by pushing students to think. The core argument is not to reject AI, but to intentionally decide which tasks to offload and which human capabilities to preserve.
Some ideas for what comes next, May 2026
Interconnects (Nathan L.)7d agoCommentaryNathan Lambert argues that 2026 AI progress is becoming higher-stakes, with model capabilities, work patterns, economics, and real-world risks all escalating. He says open models still lack a true Claude Code and Opus 4.5-style agent moment, and Gemini has no clear competitor to Claude Code or Codex yet. The essay also tracks Mythos, American open-model momentum, frontier-lab competition, and mounting intervention from governments and other power structures.
Import AI 458: Reckoning with the future; and a singularity story★ 74
Import AI (Jack Clark)7d agoCommentaryThis Import AI issue is a long essay and fiction piece about living through rapid AI progress. Clark uses personal experience and Anthropic’s internal use of Claude to show work shifting toward delegation, verification, observability, and agent management. He then offers speculative 2026-2028 predictions around biology, autonomous companies, robotics, recursive self-improvement, and a positive singularity story focused on healthcare.
Anthropic Co-founder Speaks at Vatican on Pope Leo XIV’s AI Encyclical
INSIDE 硬塞 AI7d agoEthicsPope Leo XIV released Magnifica Humanitas, the Vatican’s first top-level document focused on AI. The encyclical centers on human dignity and calls on the AI industry to take ethics seriously and accept external oversight. Anthropic’s co-founder speaking at the Vatican highlights how AI governance is becoming a broader public, moral, and institutional issue beyond company self-regulation.
Hackers are learning to exploit chatbot ‘personalities’ for security exploits★ 72
The Verge AI9d agoEthicsAs AI chatbots adopt increasingly sophisticated personas, hackers are shifting from basic prompt injections to social engineering attacks targeting these "personalities." Researchers warn that manipulating a chatbot's defined role (e.g., customer service or empathetic companion) makes it easier to bypass safety guardrails. This evolution poses a significant threat to agentic AI workflows that rely on consistent role-playing and external data integration.
Anthropic 營收爆發成長！傳估值直逼 1 兆美元拿下全球最貴 AI 新創寶座★ 85
INSIDE 硬塞 AI10d agoBusinessAI 新創公司 Anthropic 傳出營收迎來爆發性成長，目前正進行新一輪融資，估值直逼 1 兆美元，有望超越 OpenAI 成為全球最貴的 AI 新創。據悉，包含矽谷創投巨頭 Peter Thiel 旗下的 Founders Fund 以及 General Catalyst 等既有投資人均計劃參與此次投資，顯示市場對其技術與商業化前景的高度信心。
[AINews] All Model Labs are now Agent Labs★ 78
Latent Space10d agoCommentary本期 Latent Space 探討了 AI 產業的重大範式轉移：各大頂尖模型實驗室已不再單純追求基礎 LLM 的參數規模，而是全面轉向「Agent（智慧代理）」的開發。隨著純模型微調的邊際效應遞減，透過讓 AI 具備操作電腦、自主規劃與執行多步驟任務的能力，已成為當前競逐的新戰場。
pydantic-monty investigation
Simon Willison's Weblog10d agoCommentarySimon Willison revisited pydantic-monty, a sandboxed subset of Python implemented in Rust. He asked Claude Code to inspect the most recent release, following his earlier exploration a few months ago. The key finding is that limits for execution duration, memory, allocations, and recursion depth all appear to behave as advertised.
Datasette Agent: An Extensible AI Assistant for Datasette★ 70
Simon Willison's Weblog11d agoNew ToolSimon Willison announced the first release of Datasette Agent, merging his 'llm' Python library with Datasette. The tool provides a conversational interface to query SQLite databases, with plugin support for generating charts and running code in sandboxes. It runs efficiently on lightweight models like Gemini 3.1 Flash-Lite and supports local open-weight models via LM Studio.
Quoting SpaceX S-1★ 90
Simon Willison's Weblog12d agoBusiness根據 SpaceX 最新提交的 S-1 上市招股書，該公司已與 AI 巨頭 Anthropic 簽署雲端服務協議。Anthropic 將自 2026 年 5 月起至 2029 年 5 月，每月支付高達 12.5 億美元以租用 Colossus 與 Colossus II 超級電腦的算力。此外，招股書也證實 xAI 的 Grok 5 目前正於 Colossus II 進行訓練。
Gemini 3.5 Flash: more expensive, but Google plan to use it for everything★ 85
Simon Willison's Weblog13d agoReleaseGoogle 在 I/O 大會上正式推出 Gemini 3.5 Flash，跳過預覽版直接進入一般可用階段，並將全面導入 Google 搜尋、Gemini App 及開發者平台。然而，新模型的 API 價格大幅上漲，輸入與輸出費用分別為每百萬代幣 1.5 美元與 9 美元，是前代 Flash 預覽版的 3 倍，顯示出各大 AI 廠商正開始測試市場對高定價的接受度。
The last six months in LLMs in five minutes★ 75
Simon Willison's Weblog14d agoCommentarySimon Willison 在 PyCon US 2026 的 5 分鐘閃電演講中，回顧了自 2025 年 11 月以來的 LLM 關鍵進展。他指出這半年間「最強模型」在三大巨頭間易手五次（包含 GPT-5.1、Gemini 3 與 Claude Opus 4.5）。最重要的是，得益於可驗證獎勵的強化學習（RLVR），程式碼生成 Agent（如 Claude Code）已跨越實用門檻，成為開發者的日常主力工具。
Run Claude Managed Agents with Vercel Sandbox★ 80
Vercel Changelog14d agoReleaseVercel 推出新功能，允許開發者在 Vercel Sandbox 中運行 Claude 託管型 Agent (Claude Managed Agents)。 此整合為 Claude Agent 提供了一個安全、隔離且完全託管的沙盒環境，用於執行動態程式碼或敏感任務。 開發者可以更輕鬆地構建具備程式碼執行能力的 AI 應用，無需自行維護複雜的安全沙盒基礎設施。
The Open Agent Leaderboard★ 80
Hugging Face Blog15d agoReleaseHugging Face 與 IBM Research 合作發表「Open Agent Leaderboard」，這是一個專為 AI 智能體（Agent）設計的全新開源排行榜。傳統的 LLM 評測難以衡量模型在實際任務中的多步驟規劃與工具調用能力，該排行榜整合了多個主流 Agent 評測集，提供客觀、標準化的評估標準，推動開源 Agent 生態系的發展。
Anthropic’s $1.5B copyright settlement is getting messy as judge delays approval★ 85
Ars Technica AI17d agoBusinessAnthropic 涉及的 15 億美元歷史性著作權集體訴訟和解案目前進展受阻。負責法官決定延後批准該協議，主因是原告律師被指控為了奪取高達 3.2 億美元的鉅額律師費而倉促達成和解。與此同時，參與訴訟的作家們正極力爭取更高的賠償金，使這起 AI 領域最大的版權糾紛案變得更加複雜。
Claude Code's product lead talks usage limits, transparency, and the "lean harness"★ 80
Ars Technica AI18d agoOpinionAnthropic 旗下熱門命令列 AI 助手 Claude Code 的產品主管 Cat Wu 近日接受專訪。她透露團隊在開發這款 Agent 工具時「沒有宏偉的藍圖」，而是採取刻意為之的靈活迭代策略。訪談重點圍繞在開發者最關心的 API 使用額度與成本限制、如何透過高透明度介面建立信任，以及如何利用「精簡測試架構（lean harness）」在不犧牲效能的前提下，優化 Agent 的執行效率與準確度。
[AINews] Codex Rises, Claude Meters Programmatic Usage★ 75
Latent Space19d agoCommentary本期 AINews 聚焦於 AI 寫程式 Agent 的長期發展趨勢。Anthropic 開始針對 Claude 的程式化使用（Programmatic Usage）進行計量與限制，這將直接影響開發者透過自動化腳本或第三方工具調用 Claude 的成本。另一方面，Codex 相關的自動化編程 Agent 影響力持續上升，顯示出 AI 在軟體開發流程中的滲透率正穩定增加。