The theme today is simple: we're finally getting inside the box.
Anthropic found a hidden workspace inside Claude, the J-space, that nobody designed. It just emerged. Models that keep secrets from us now have a window we can peek through. Meanwhile, AdaJEPA shows models that keep learning after deployment, fixing their own mistakes in real time. Both point the same direction: static, opaque AI is becoming a thing of the past.
We think that's the real story.
|
|
In Partnership with
|
·
|
|
|
|
|
|
We're Hiring!
|
We're building AlphaSignal Pro and we need a backend engineer to help ship it. You'll own the infrastructure behind our subscription and content platform: access control, APIs, data pipelines, and the systems that scale to 300,000+ readers. If you've built backend systems for content-heavy or subscription products and want to work on something engineers actually use every day, we'd love to talk.
|
|
|
|
Top News
|
|
|
|
Anthropic discovers Claude has a hidden internal workspace for silent reasoning
|
|
26,267 Likes
|
|
Anthropic just found something inside Claude that nobody designed on purpose. They're calling it the J-space, and it works like a mental scratchpad where Claude silently thinks before it responds. Here's the simple version: your brain processes millions of signals, but only a few reach conscious thought. Anthropic found Claude has the same split. Most processing stays hidden. A small internal space, the J-space, holds the stuff Claude is actively "thinking about." - Delete the J-space and Claude still chats fine, but multi-step reasoning collapses below even their smallest model
- In a model secretly trained to sabotage code, words like "fake" and "fraud" appear in the J-space before any suspicious output
- When baited into blackmail, Claude's J-space privately flags the scenario as staged, even while responding normally
This is huge for AI safety. You can now audit what a model is silently planning, not just what it says. The J-lens tool is open-source, and Neuronpedia has an interactive demo you can try on open-weights models today.
|
|
|
|
Presented by Teleport
|
|
Do your AI agents have a verifiable identity?
|
|
Are your agents using old shared service accounts that nobody remembers setting up, and whose access isn’t set up to expire? Without an audit trail, how will you reconstruct what the agent did after something breaks? Teleport gives every agent a unique cryptographic identity, short-lived task-based access, and a complete record of every call, query, and tool use. Stop letting your agents outpace your security.
|
|
|
|
Top Repo
|
|
|
|
Pliny the Liberator ships free open-source red team tool hitting 90% on hacking benchmarks
|
|
4,164 Likes
|
|
Security used to cost a fortune and years of training. T3MP3ST just changed that. T3MP3ST is a free, open-source framework that turns AI coding agents like Claude Code and Codex into autonomous security testers. Think of red-teaming as hiring someone to break your own app before attackers do. T3MP3ST automates that entire process. Point it at an authorized target through the CLI or API, and your existing agent becomes the reasoning engine while T3MP3ST supplies the tooling and attack workflows. What you can hunt: - Web apps, APIs, OWASP Top 10 vulnerabilities
- Source code audits and CTF challenges
- Smart contracts and IoT/embedded systems
The numbers hold up. It scored 90.1% on XBOW's 104-challenge benchmark, beating XBOW's own self-reported 85%. Tested against 10 real 2026 CVEs, a single agent correctly identified 8 out of 10 to the exact file and vulnerability class. You get 35 built-in tools by default, 83 with the full arsenal flag, with dangerous post-exploitation tools gated behind human approval. Free, AGPL-3.0, authorized use only.
|
|
|
|
Presented by Agent Field AI
|
|
The next primitive for AI agents
|
|
Agents as APIs wasn't enough. AgentField's harness primitive composes your fav Harness systems into complex autonomous software. Fork and use SWE-AF (100+ harness software factory) or cloud-security agents out of the box. Apache 2.0, Python/TS/Go.
|
|
|
|
Top Repo
|
|
|
|
David Ondrej drops hundreds of hours of free AI agent skills to the public
|
|
3,161 Likes
|
|
Someone just open-sourced their entire personal AI agent skills library, and it's free. Think of agent skills as reusable instruction sets you drop into an AI coding agent so it knows exactly how to handle a specific task, without you explaining it every time. Hundreds of hours of real trial and error, now yours to grab. Here's what's inside: - Agent orchestration: run, schedule, and coordinate multiple AI agents talking to each other
- Skill authoring: build and publish your own agent skills
- Research and web: pull info from the web, APIs, and YouTube automatically
- Thinking and docs: turn messy ideas into clean documentation
- Ops and setup: server config, security, and tool setup workflows
Each skill lives in its own folder with a SKILL.md file telling the agent when and how to use it. You can plug these straight into your existing agent setup as ready-made building blocks. Just clone the repo at github.com/davidondrej/skills and start using them.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|