Pinned Tweet
Verifying my account on nostr
My Public Key: "f0293508f3eb9e6fe99c2fd8ba69ff446216872a2d9f67979bfa4db8b3155806"
Find others at nostr.directory @nostrdirectory #nostr
4
636
w3ird retweeted
Today, we launch Tinfoil Bunkers.
More of a fortress than a sandbox, Bunkers are confidential workspaces in the cloud. They're made for your agent's most sensitive, longest-running jobs. Every Bunker runs in a hardware enclave, so you can verify that not even Tinfoil can look inside. You can connect your accounts and the tools you use without handing your agent the keys, and you decide how far those keys reach.
Since the invention of personal computing, you’ve had to choose to have your work happen either on your machine or in the cloud. Your laptop gives you privacy, but your agents stop when you close it. Most clouds stay on, but whoever runs them can see inside. We designed Bunkers so you don't have to choose one or the other.
Try spinning up a Bunker for your agent to triage your inbox. Or give your agent a jackhammer. Whichever you pick, we won't know. 🙂
tinfoil.sh/bunkers
w3ird retweeted
It is with extreme sadness that we share with you the devastating news that Fiona did not make it after multiple attempts from our amazing vets and keeper team to help her through what ended up being a medical emergency. cincinnatizoo.org/beloved-hi…
w3ird retweeted
‼️ BREAKING: Researchers broke past PHP's fix for the php2shell flaw and can again run their own code on web servers, no login needed.
Security firm depthfirst says the latest PHP releases are still vulnerable. The bug, CVE-2026-126714, is rated 9.2 out of 10.
depthfirst.com/research/phpa…
Readers added context they thought people might want to know
PHP 8.4.28 and 8.5.13 fixes CVE-2026-126714.
php.net/ChangeLog-8.ph…
php.net/ChangeLog-8.ph…
w3ird retweeted
Pre-auth RCE in nginx is the bug every hacker dreams of finding. It almost never happens.
Today we’re disclosing nginx2root: five nginx vulnerabilities, and an exploit that takes an unauthenticated attacker all the way to root.
One of themhits the default config.
If you run nginx, patch now. Then read: nginx2root.com
CVE-2026-123007, CVE-2026-123634, CVE-2026-123633, CVE-2026-121810, CVE-2026-113425
w3ird retweeted
Fellow researchers: if you're using AI, you should be doing so privately.
Why send your research data & intellectual property to private AI companies?
"no training" & similar promises don't go nearly far enough.
Good news: in ~ 5 minutes, you can be using AI with the ease of ChatGPT but much better privacy.
Easy Mode: AI in Encrypted Enclaves
Today, companies like @TinfoilAI and @TryMapleAI serve open-weight AI models inside encrypted enclaves.
Using these services is a huge level-up: they don't train on your prompts. In fact, they can't even see them. Risk of your prompts getting exposed from a breach goes way down, too.
No risk goes to zero, but this is a big, meaningful improvement. Just a handful of minutes to get started.
Bigger Picture: Bring Inference Back To the University
Universities should be hosting open-weight models, fine-tuning them, and making them dead-easy for researchers to work with.
Ideally, universities are serving this inference in encrypted enclaves & developing research lines on how to do it effectively.
Huge leadership potential & public benefits for academic and intellectual freedom, and open knowledge.
Want to dip your toes in encrypted enclaves? @micahflee has a nice recent piece:
micahflee.com/heres-how-anyo…
w3ird retweeted
Inkling 2 Small is now in Maple.
@thinkymachines' open-weight model is built for long work. In Maple Work, we had it:
→ turn a year of business statements into a P&L and flag what's off
→ build a 3D maple forest from one prompt
Connect Square, Stripe and more. Privately.
so i can have double the bad takes
Introducing Second, and a $10M seed led by @a16z.
Second builds a model of you that you own, and your agents can use as an API.
Truly personal software needs a deep understanding of you: not just your emails and messages, but how you work, what motivates you, your people, priorities, style, goals, and intent.
This requires a new kind of personal AI infrastructure. One that continuously aggregates your context, distills it into durable understanding with a ton of compute, and then serves it securely, ubiquitously, and up-to-the-minute. One that builds your *portable user model*.
Second is that infrastructure. It installs seamlessly on your Mac, and your context just comes out. Live APIs of your priorities, to-dos, wikis, writing samples, preferences. Or what you did minutes ago. Or context packs of anything, dynamically assembled in seconds.
Second is streaming-state from local to a private cloud, which means anything you connect to--whether Claude, Dot, Muse, Grok Bot, Instinct, or iMessage--shares the same, growing, continuous understanding of you.
Your portable user model is as powerful as it is sensitive. We founded @FirstPersonLabs on the principle that it should belong to you. This commitment means taking on hard systems and security problems and making sovereignty an invariant of Second from the beginning.
We've barely scratched the surface of what portable user models can unlock. Deep understanding of us, served at machine speeds, could drive the next leap in personal software, human-computer interaction, large-scale coordination of humans and agents, and more!
We're tremendously grateful to be supported by @a16z (@martin_casado, @YODELAGARZA), Mozilla Ventures, @JeffDean, Ram Shriram, Jeetu Patel (@jpatel41), @percyliang, J. Alex Halderman (@jhalderm), @sanmikoyejo, and many others.
Our vision is for a future where every person is more powerful, because they have the ultimate affordance: personal computers that fully understands them on their own terms.
If this excites you, we're hiring and come build with us!
1
4
214
w3ird retweeted
We read the ZDR docs of 12 AI providers.
"Zero" means something different for everyone. For Tinfoil, zero is something you can verify. Can you trust ZDR policies when they are riddled with exceptions and caveats?
tinfoil.sh/blog/2026-11-16-z…
w3ird retweeted
We break: The highly classified network that the National Security Agency uses for its internal communications is down, according to five people familiar with the matter.
The network, called NSA Net, allows employees to communicate classified information internally. At this point there is no evidence of a cyberattack, sources said.
As the agency works to understand the scope of the problem, it has enacted staggered arrivals for personnel, the sources said.
CNN has asked the National Security Agency for comment.
cnn.com/2026/11/17/politics/…
Packages for StartOS of rbitcoin, rbitcoin-space, and a modified version of public-pool (which allows switching between Core and rbitcoin as backend) are on my Github and startos registry, add registry.testnet4.info
rbitcoin.space v0.3.0 is out.
- MCP support is out, rbitcoin.space/mcp ask grok bot or any MCP client to add this.
- Via the MCP you can decode invoices, look up txs, decode raw bitcoin txs, ask about an ark VTXO and much more
- Decode raw bitcoin txs and broadcast them from the UI after previewing them.
w3ird retweeted
I built rbitcoin.space a rewrite of @mempool in Rust and SolidJS for being very fast and lite weight backed by @rbitcoin_org and @lightningdevkit
Right now running Mainnet, Signet, Testnet4 and Testnet5.
The first goal is feature parity with mempool space so everything you’re used to is already there and we can build further on top of it.
- Significantly less resources needed to self host and also way less disk space used.
- it’s fast
- Way less things to run, just rbitcoin and rbitcoin.space. No electrum, no Bitcoin core, no LND/CLN, no Maria DB.
- runs a lite weight LDK node just for gossip
- most work is done by the rbitcoin node itself, the storage layer is now a very simple SQLite.
It’s running on my home server. Please don’t crash it! 💀 I’ll move to cloud hosting eventually.
w3ird retweeted
HMRC want the power to snoop on your conversations with AI.
This is, quite frankly, absolutely none of their business.
I strongly encourage everyone in the UK to look into @TryMapleAI and other privacy-preserving AI alternatives (whether local models or otherwise) in order to head off this madness at the pass.
w3ird retweeted
We've been cooking on our new Maple Work agent for awhile, so happy to have it out there. A ton went into this that deserves their own highlights, but briefly:
- TEE models built in, no configurations needed
- Remote first: An enclave assisted remote pairing option to all your computers, encrypted comms through IROH relays. I did everything I could to make remote feel as smooth as local.
- Pi inspired: A rust clone of the architecture, will be compatible with Pi plugins and extensions in the future
- GPUI based, because of course.
- @trycua driver built in, works for all platforms
- ACP: works with harnesses such as paseo, works both ways too, can even use Codex and Claude as a subagent
And what I'm looking forward to that's not out yet:
TEE based sandboxes, which will pair well with our remote first design.
i'm at the point now that if i can't find a bug in some code using an abliterated model, i don't trust the code.
1
2
341
w3ird retweeted
>Telegram releases a widget "sandbox" for untrusted js
>ask durov if there is actual process isolation
>he doesn't understand
>explain that a sandbox should be its own process
>he laughs and says "it's a good sandbox sir"
>write javascript exploit
>zero isolation
⚠️⚠️ All LNbits users should stop their LNbits servers IMMEDIATELY. ⚠️⚠️
Please forward this announcement.
Following a wave of attacks, we have a number of security patches to push. Until then, it's better to err on the side of caution. 🙏
Updates to follow soon.
w3ird retweeted
Built this for my own company like 6 months ago. Down to the nix and multi-agent-single-computer setup.
Except it’s BYOC and on prem capable, because renting persistent mac/win/linux boxes for every dev is very expensive versus buying physical hardware.
Every company should own their secure cloud dev environment where agents can change code, run the full application, and verify their work. At @depthfirstlabs we care tremendously about this and today we share how we built our environment.
Each developer gets a persistent devbox, and each task gets its own application stack, database, and workflow state, so agents work in parallel. Engineers have flexibility to switch between Codex, Claude Code, and other agents without losing context.
The environment runs inside our private AWS network, with access controlled through @Tailscale and package requests checked by depthfirst’s Dependency Firewall in configured build and registry paths.
When an agent finishes an iteration, it can share reproduction steps, video evidence, and a private preview URL. We can open the exact environment it tested and review the behavior ourselves.
We had it operational within a week.
plot twist: they just made it watch team america over and over
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
3
10
534





























