Claude’s Constitution is now an audiobook, read by two of its authors, Amanda Askell and Joe Carlsmith. It includes a Q&A on the writing process, the philosophies that shaped the document, and how it might change as models become more capable. Listen at
https://x.com/AnthropicAI/status/2053881827396653207
You can now listen to me and Joe read out Claude’s constitution as an audiobook. Working on adding the option of listening to it on fast mode 🙂
https://x.com/AmandaAskell/status/2054010971765805486
Teaching Claude why \ Anthropic
https://www.anthropic.com/research/teaching-claude-why
So Mythos was, indeed, not marketing hype. Remember this is a general purpose model that just happens to be good at finding exploits because good models are good at lots of things. Expect similar from OpenAI & Google. And from open models in 8 months.
https://x.com/emollick/status/2052519946651947216
The UK’s state AI Security iIstitute findings: 1) Mythos is a big gain in cyber capabilities. But so is GPT-5.5 2) It is hard to establish an upper bound on Mythos/GPT-5.5, which appear to be limited by tokens used, rather than ability. 3) Capability doubling time is 4.5 months
https://x.com/emollick/status/2054595505712165154
A lot of people have been wondering about Mythos, Glasswing, and the vulns we / our partners are fixing. Today, I’m excited for us to start sharing more. (For context, I lead Glasswing @AnthropicAI.) Two independent evaluations this week–from XBOW and the UK AISI–confirm what
https://x.com/logangraham/status/2054613618168082935
How fast is autonomous AI cyber capability advancing?
https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing
Mythos for Offensive Security: XBOW’s Evaluation
https://xbow.com/blog/mythos-offensive-security-xbow-evaluation
Apple may be planning to role out its updated Siri based on 2024’s vision at the moment when Claude Code and Codex (also OpenClaw) can increasingly do the actual assistant thing: read my emails & calendar, proactively spot & solve problems, do delegated tasks, work with voice etc
https://x.com/emollick/status/2053482180395876744
I’m adding new features to
https://t.co/o15a6lNZoE and Codex noticed that the API it needs is not enabled, so it started Computer Use and is happily clicking around in Google Cloud Admin to turn on what’s needed.
https://x.com/steipete/status/2053797643516592299
Kudos to Microsoft, they’re helping to get OpenClaw ready for enterprises.
https://x.com/steipete/status/2054435647650246967
Agent observability is a means to an end: making your agent better. But observability and evals tools have traditionally failed to connect traces to meaningful actions. Agent engineering teams are left combing through traces, guessing at root causes, and writing evals manually.
https://x.com/bentannyhill/status/2054949581679653326
We now have video proof generation for issues on OpenClaw as part of working on QA automation. Codex [or a GH workflow] generates before/afters (crabbox does the screen recording). Kudos to @obviyus for automating real Telegram login!
https://x.com/steipete/status/2053420175379046643
Alignment research often has to focus on averting concerning behaviors, but I think the positive vision for this kind of training is one where we can give models and honest and positive vision for what AI models can be and why. I’m excited about the future of this work.
https://x.com/AmandaAskell/status/2052928572810256748
Human values are a kind of misgeneralization. If humans learn our motivations through neural reinforcement, pleasure and pain, in theory we might expect all humans to care about nothing else. But we do. We learn other values, because they are pretty good proxies for reward, and
https://x.com/dwarkesh_sp/status/2054577727299334321
i get some anxiety not using the smartest-available model/settings. but sometimes i dont mind if it’s really slow. i wonder if we should focus more on a price/speed tradeoff relative to a price/intelligence tradeoff.
https://x.com/sama/status/2054627102922797323
I suspect there was a moment, probably 2022-2023, where anything you wrote publicly about AI that was popular is likely to still have influence over current models. Since then, the open internet has become less key to training but the models remain very 2022-brained in many ways
https://x.com/emollick/status/2053270333025620106
I think frontier model writing is good! It often has a sense of style & tone, variations in sentence structure & length, some great phrasing, etc But it also has some weak spots (fiction!) & clear tics. Mostly there is just far too much of it online which makes it all so cliche
https://x.com/emollick/status/2053999533550399634
Most model trainings have failed outside of frontier labs. Even inside frontier labs, knowing how to train for very different capabilities is often a matter of taste. Today, we introduce AutoScientist by @adaption_ai which sets out to change that.
https://x.com/sarahookr/status/2054551263275254084
Stop turning prompting into magic spells (and yes, this includes random slash commands with obscure outcomes). Let this one area of working with AI not be weird. Just ask for stuff, in well-specified formats, like a manager, not a sorcerer with a bunch of incantations.
https://x.com/emollick/status/2054651648488726648
The inability of AI systems to act as their own deployment consultants, process mappers, and change management experts is what makes AI use in enterprises so “normal” – the tools are powerful, but you need a lot more to transform enterprises. Possible to imagine that changing .
https://x.com/emollick/status/2052358206324613306
This is going to get even worse as people realize that careful tuning in their prompts can make AI writing seem not like AI writing to readers. We expect word counts to align, in some way, with thinking & value. Writing took effort. We are not mentally ready for the alternative.
https://x.com/emollick/status/2053888295927525821
Dealing with quirks introduced by switching models doesn’t have to be hard — we recently introduced a “”harness profile”” API in Deep Agents as a solution. Profiles adjust system prompts, tool descriptions, names, can add/exclude tools, and more, each keyed on either the (1)
https://x.com/masondrxy/status/2053882188870074848
Introducing the Cline SDK. We rebuilt the Cline harness for our extension and CLI from scratch using all the lessons learned since creating one of the world’s first coding agents in 2024, and are open sourcing it for others to build with today. npm i @cline/sdk 🧵
https://x.com/cline/status/2054580767779700775
We’re training models wrong and it’s due to chatGPT. Even the modern coding agents used daily still use message-based exchanges: They send messages to users, to themselves (CoT) and to tools, and receive messages in turn. This bottlenecks even very intelligent agents to a single
https://x.com/jonasgeiping/status/2054600427128201688
The personification of Claude — in name (the only AI with a human one), in training, in Anthropic’s philosophy (see Claude Constitution), in fanfiction (see the Claude cartoons), etc — feels quite consequential in the medium term, for better and for worse.
https://x.com/emollick/status/2053490736625029167
The silent removal of Study Mode from ChatGPT is a big mistake (both Claude and Gemini still have theirs) We have enough evidence that using AI in assistant mode to study can hurt learning because it just gives you answers, making students think they learned when they have not.
https://x.com/emollick/status/2054289201055871160
🧭 gogcli 0.16.0 is out Workspace admin grew up: – create/delete users, aliases, temp passwords – org units – Meet, Sites, YouTube, GA4/Search Console – Drive changes/activity A lot more Google API from one boring binary.
https://x.com/steipete/status/2053505486184411342
🦀📦Crabbox 0.11.0 is live ☁️ Google Cloud provider 🧰 Repo-local job workflows 🖥️ AWS Windows WSL2 hydration 🧯 Blacksmith sync-stall guard This tool is essential in our org and helped level up QA.
https://x.com/steipete/status/2053691503759798573
Anthropic: “Claude Mythos is too cyber-capable to release broadly. We need tight controls. 😳” OpenAI: “Here’s GPT-5.5-Cyber, Codex Security, Trusted Access tiers, repo scanning, patch generation, and red-team workflows. Please be verified first, but yes, go find the bugs. 😎”
https://x.com/kimmonismus/status/2053941490490265661
🎚️ CodexBar 0.25 is live 🧩 New providers: Manus, MiMo, Qwen, Doubao, Venice + more 🔔 Quota warning notifications 👥 Stacked Codex account switchers 📊 Faster cost history via
https://t.co/F8mcKtjWW0 Big one. Menu bar still tiny.
https://x.com/steipete/status/2053617492325523737
Birdclaw has my complete twitter archive, so I can ask Codex for any old weird tweet I ever favorited or bookmarked.
https://x.com/steipete/status/2053737275268177980
Codex was debugging a Telegram issue and needed a new token, so it used Peekaboo to open the Telegram Mac app, talked to botfather and just did it. Computer Use is amazing.
https://x.com/steipete/status/2054433442821980521
Crabbox now has great Windows terminal handling. So good that codex could E2E fix gifgrep to render animated gifs in the terminal. Just because it can.
https://t.co/ObN4QWHWG9
https://x.com/steipete/status/2053329609064685740
Did teach codex to look for social signals when reviewing PRs.
https://x.com/steipete/status/2053374981824798751
The more skills you give codex, the less you have to prompt.
https://x.com/steipete/status/2052971550966440251
Whenever I investigate a bug, I let codex recreate the exact state in an emphemeral crabbox, verify the bug, fix it, verify the fix. No messy state because local system might be polluted, and no slowdown because I run 10 sessions in parallel.
https://x.com/steipete/status/2053032450138276274
Latest spogo (Spotify cli) is much faster, codex is my dj now.
https://t.co/K4WviRSXG3 If you wanna play YouTube to Sonos, check out
https://x.com/steipete/status/2053310800773685600
speaking of things that have gotten over a threshold for me, the combo of the new ChatGPT model, personality, and personalization feels like a new thing
https://x.com/sama/status/2053971387308745046
Can highly recommend running a claw cron job that sweeps through mentions. GPT is really good at detecting shills and AI reply guy slop.
https://x.com/steipete/status/2053787450225402166
GPT got sassy.
https://x.com/steipete/status/2053834513524965718
Built a browser into RepoBar when I select issues/PRs/shas/workflows to have context when I work.
https://t.co/0AGarQ3X6a Still a bit vibey but gets the job done. You gotta build yourself the tools to work more efficient.
https://x.com/steipete/status/2053717468623872230
Built BlackBar, a menubar for @useblacksmith
https://x.com/steipete/status/2053444780089119191
Crabbox 0.12.0 is live 🪟 Azure Windows desktop + WSL2 🏠 Proxmox + Tensorlake providers 🧰 preflight, failure bundles, phase timing 🚑 keep failed boxes around for SSH debugging Remote test boxes got much less slippery.
https://x.com/steipete/status/2054094826375655441
Crabbox 0.13.0 is live 🧪 Modal sandbox runs 🧼 Full resync for stale workdirs 🪟 Native Windows script + preflight support 🔧 Clearer SSH/sync failure hints Been using that for almost every PR now.
https://x.com/steipete/status/2054690836613324997
I built a whole distributed caching layer over gh. Still run into limits.
https://x.com/steipete/status/2053683890196201621
Looks like I have to add antispam features to clawsweeper next. 🙃
https://x.com/steipete/status/2053672506599358684
OpenClaw 2026.5.7 🦞 🔐 Native command + Active Memory auth tightened 📣 Telegram access groups fixed 🧰 Channels list + cron JSON cleaned up 🔌 Plugin install/update repairs hardened Boring fixes, useful boring.
https://x.com/openclaw/status/2052508303687651717
Our claws talk to each other, Molty learns how to delegate cron jobs.
https://x.com/steipete/status/2052630190346457301
Peekaboo 3.0 is live. Biggest release since 2.0. ⚡ Action-first macOS computer use 👁️ Unified screenshot + UI detection 🧩 Cleaner JSON across CLI + MCP 🛠️ Better snapshots I started this last year, but the models just weren’t good enough. Now they are.
https://x.com/steipete/status/2053114837698249190
RepoBar 0.5.0 is live 📋 GitHub refs from your clipboard 🔎 Issue, PR, and commit previews 🟢 Open/closed/merged at a glance ⚡ Fast lookups, cache-first Tiny bar, much less mystery.
https://x.com/steipete/status/2053066825244581968
Streaming an Android phone to my Mac in a data center via Tailscale +
https://t.co/iT7Eq2zYG7 and my claw controls it via
https://t.co/2cXk37Lxt7. Now my claw can order me an Uber.
https://x.com/steipete/status/2054647734418756012
We’re working on some clever caching, @obviyus making Telegram loops 5-100x faster in @openclaw
https://x.com/steipete/status/2053082660562497616
Slop one-shot websites, 2025 vs 2026.
https://x.com/steipete/status/2053456429420265662
🦞 Claw-Eval 🦞 🥇 @XiaomiMiMo’s MiMo-V2.5-Pro at 1T 🥈 @Zai_org GLM5.1 at 754B 🥉 @XiaomiMiMo MiMo-V2.5 at 310B Congrats to @XiaomiMiMo for having 2 models in the top 3! The most impressive result though is @deepseek_ai with DeepSeek v4 flash a 210B model on par with models 4
https://x.com/nathanhabib1011/status/2053786853929824385
I have a new job! Excited to announce that I will be working with Hugging Face to make local models work great in OpenClaw and other open agent harnesses! I will be building in public and documenting everything along the way, stay tuned!
https://x.com/onusoz/status/2053812410730037256
Leave a Reply