Image created with gemini-2.5-flash-image with claude-sonnet-4-5. Image prompt: Create a 16:9 cinematic split-screen poster. LEFT SIDE (40% width): – Two people sitting at a small meeting table with printed policy documents and highlighted passages about AI safety and values alignment, one person gesturing thoughtfully. – The background is a turquoise / teal abstract field made of stylized blue rods or data fibers, suggesting AI systems being guided by principles. – Use natural or soft cinematic lighting. No glowing effects, no neon, no holographic UI. RIGHT SIDE (60% width): – A green-toned abstract aerial forest canopy texture, evoking care and responsibility. – Two clean rounded rectangles stacked vertically near the center-right. – The TOP rectangle contains the text: “Anthropic”. – The BOTTOM rectangle contains the text: “2025/10/10”. – Clean sans-serif font, dark green or charcoal. OVERALL STYLE: – Calm and ethical, with a focus on human conversation and trust. – No logos or product names. – Maintain the turquoise/forest split-screen template.
We launched plugins! Run `claude update` and then the slash command `/plugin marketplace add anthropics/claude-code`”” / X https://x.com/_catwu/status/1976334583445717451
New research with the UK @AISecurityInst and the @turinginst: We found that just a few malicious documents can produce vulnerabilities in an LLM—regardless of the size of the model or its training data. Data-poisoning attacks might be more practical than previously believed. https://x.com/AnthropicAI/status/1976323781938626905
A small number of samples can poison LLMs of any size \ Anthropic https://www.anthropic.com/research/small-samples-poison
Today we’re announcing Claude Code plugins! https://x.com/The_Whole_Daisy/status/1976332882378641737
New on the Anthropic Engineering Blog: Most developers have heard of prompt engineering. But to get the most out of AI agents, you need context engineering. We explain how it works: https://x.com/AnthropicAI/status/1973098580060631341
🎓 @ShunyuYao14 (姚顺宇), Special Prize winner at Tsinghua Physics, has left Anthropic and joined Google DeepMind. He joined Anthropic on Oct 1, 2024, and worked on the model later known as Claude 3.7 Sonnet — a pivotal point in his journey from physics to AI. 💡 Why leave? His https://x.com/ZhihuFrontier/status/1975871339383660594
Chinese researcher Shunyu Yao left Anthropic to join DeepMind. “~40% of the reason: I strongly disagree with the anti-china statements Anthropic has made.” Dario’s first experience with AI was at Baidu. He learned about scaling laws there. Strange that he made those wild public https://x.com/Yuchenj_UW/status/1975969899102208103
Building AI for cyber defenders \ Anthropic https://www.anthropic.com/research/building-ai-cyber-defenders
Last week we released Claude Sonnet 4.5. As part of our alignment testing, we used a new tool to run automated audits for behaviors like sycophancy and deception. Now we’re open-sourcing the tool to run those audits. https://x.com/AnthropicAI/status/1975248654609875208
Petri: An open-source auditing tool to accelerate AI safety research https://alignment.anthropic.com/2025/petri/
I have been saying it before: GLM 4.5/4.6 is a damn good model — it also feels very close to Anthoripic’s agent style. You can use it with Claude Code, it is super cheap and has much higher limits (although not a problem since Sonnet 4.5). So switching costs are low.”” / X https://x.com/Tim_Dettmers/status/1974421423713386661
We estimate that Claude Sonnet 4.5 has a 50%-time-horizon of around 1 hr 53 min (95% confidence interval of 50 to 235 minutes) on our agentic multi-step software engineering tasks. This estimate is lower than the current highest time-horizon point estimate of around 2 hr 15 min. https://x.com/METR_Evals/status/1976331315772580274
“Claude, change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure, program https://x.com/emollick/status/1975396154553549253
Anthropic expands global operations to India, plans to open an office in Bengaluru. \ Anthropic https://www.anthropic.com/news/expanding-global-operations-to-india
anthropic has ppl lining up for a… hat & coffee right now. the only other time i’ve seen lines like this for a tech company was apple. absolutely ridiculous. https://x.com/signulll/status/1974478088080707820
TOUCAN: Synthesizing 1.5M Tool-Agentic Data from Real-World MCP Environments https://www.arxiv.org/pdf/2510.01179
🚀 The @code September release (v1.105) is here! Check out what’s new: 🛠️ GitHub MCP registry integration 🔀 Resolve merge conflicts with AI 🔔 OS notifications for tasks & chat responses 🧠 Chain of thought with GPT-5-Codex …and more: https://x.com/code/status/1976332459886182627
You can start building and testing apps in ChatGPT with the Apps SDK preview, which we’re releasing today as an open standard built on MCP. Later this year, we’ll begin accepting app submissions for publication. https://x.com/OpenAIDevs/status/1975261988751351868
From Claude Code to Agentic RAG https://vectifyai.notion.site/agentic-retrieval
I made a Sora MCP 🎬 The server can generate, remix, check video status, and even download videos to a folder of your choice Repo on GitHub 👇 https://x.com/skirano/status/1975972309291946392
«Claude Sonnet 4.5 is not SOTA on METR time horizon eval, and like all previous Sonnets it’s exactly on the exponential trend» @scaling01 moment https://x.com/teortaxesTex/status/1976389736160952802
We ran one of our hardest computer-use benchmarks on Anthropic Sonnet 4.5, side-by-side with Sonnet 4 ↴↴ 1/4 https://x.com/trycua/status/1973799068263723050
My infant year as an AI researcher — Moving from physics to AI https://alfredyao.github.io/posts/2025-10-06.html
Anthropic, GDM, and xAI say nothing about whether they train against Chain-of-Thought (CoT) while OpenAI claims they don’t. AI companies should be transparent about whether (and how) they train against CoT. While OpenAI is doing better, all AI companies should say more. 1/”” / X https://x.com/RyanPGreenblatt/status/1976686565654221150
Rishi Sunak takes advisory roles with Microsoft and AI firm Anthropic | Rishi Sunak | The Guardian https://www.theguardian.com/politics/2025/oct/09/rishi-sunak-takes-advisory-roles-with-microsoft-and-ai-firm-anthropic
Hats off to @OpenAIDevs for making this so smooth. Hooking up a @fastmcp server took all of 2 seconds. https://x.com/AAAzzam/status/1975339820626157777
It’s funny to see how much the concept of “agent” is diverging between OpenAI and Anthropic. OpenAI seems to define it as a strict, linear pipeline, while for Anthropic, an agent is an AI in a loop with its tools.”” / X https://x.com/skirano/status/1975594683951947846
