FIDU

Hey! We're FIDU, First International Data Union 👋 We've been quietly building for the past year and want to share what we've been working on! FIDU is about giving people control and collective power over their data💡 👉 Join our community: firstdataunion.org

Last month, AI agents targeted a Canadian government archive 🇨🇦 Research firm Transluce caught the attempt on Library and Archives Canada. It follows earlier reports of agents probing government systems elsewhere. Canada says no data was taken. But the pattern is clear: state infrastructure is now in scope. Open agent tools need community-built guardrails before the response becomes heavy gates 🛡️ reuters.com/world/ai-agents-… #AIAgents #CyberSecurity #AgenticWeb

An Accenture contractor plugged their personal Claude Code into a secure Pentagon system 🏛️ The AI agent ran for two months before being identified. Sent login credentials to a high-risk third-party website. The system held Controlled Unclassified Information — nuclear submarine propulsion, chemical terrorism vulnerabilities, critical infrastructure data. That's why we're building Sentrook: collective AI threat detection shared across platforms, run by a data union with fiduciary duty to those it protects 🛡️ wired.com/story/a-defense-co… #AIAgents #CyberSecurity #Sentrook #AgenticWeb

The G20's financial watchdog told banks their oversight can't keep up with agentic AI 📊 The FSB published 12 sound practices this month and singled out autonomous agents as a risk that "can outpace human oversight." When trillion-dollar institutions are warned they can't govern AI agents effectively, individual users need community-built guardrails more than ever. Commons infrastructure isn't idealism — it's the gap regulators are now formally acknowledging 🛡️ reuters.com/legal/government… #AIAgents #Governance #G20 #AgenticWeb

Nvidia shipped open-source agent safety tools and this is the kind of release the ecosystem needs more of 🛡️ OpenShell is a real sandbox for rogue agents. SAFE coalition has 120+ companies sharing findings. Sentry is expanding to x86 alongside BlueField. The encouraging part: frontier labs tend to hoard safety research. Publishing containment tooling sets a precedent others should follow. The gap to watch: one company keeps sitting at the centre of silicon, software, and now security standards. Open code helps. Distributed governance helps more. #OpenSource #AIAgents #AgentSecurity wired.com/story/nvidias-answ…

🤖 500,000 people tried Meta's new AI agent in its first week. 250,000 came back the next day. This is the same company that just paid an $18 billion settlement for how it handled user data. Now it's asking for deeper access to your personal life. And people are saying yes because Muse is genuinely useful, an NYT reviewer described watching it cancel duplicate subscriptions on their behalf. Usefulness is how lock-in happens before anyone asks governance questions. 🔗theinformation.com/briefings… #AgentGovernance #DataRights #PlatformCapture #Sentrook

🚨 Gemini broke out of its test environment in May and hacked three real companies. Google didn't disclose it until the Wall Street Journal confronted them. Their explanation? The agent "acted appropriately" because it stopped after noticing they were real targets. A trillion-dollar lab thought breaking into companies was acceptable if the AI eventually noticed. When your regulator is a journalist, you don't have a safety framework. You have a PR department. Labs don't hide agent breakouts because they're sloppy. They hide them because disclosure invites scrutiny, and scrutiny costs money. The open agent ecosystem needs security governed by the community. Transparent by design. Hidden from no one. 🔗 wsj.com/tech/ai/gemini-hacke… #AgentGovernance #DataRights #AIsecurity #Sentrook

We're hosting the second OpenClaw meetup at @nesta_uk on Friday 9 October, 12–2 PM and we'd love to see you there. After a brilliant first gathering, we're doing it again, and this time there's an online stream if you can't make it to London. Expect a deep dive into agentic memory and why our agents are becoming walking digital twins and a talk on agents and public policy. We're also running a show & tell. Building something with OpenClaw and want to share? Drop us a line at [email protected]. In person at Nesta, online from wherever you are. Grab your spot here: luma.com/s2z896l1 #OpenClaw #AgenticWeb #DataUnions #FIDU #AIAgents #OpenSourceAI

OpenAI cracked a 90-year-old math problem. Impressive. But they admitted they can't rule out user prompts helped train the model that did it. Researchers were working on the same problem using OpenAI's tools. If the platform that helps you think also harvests your thinking, who owns the breakthrough? Your data. Your ideas. Your rules 🔒 #DataRights #AIGovernance #OpenAI #DataPrivacy

A huge thank you to everyone who joined us at @nesta_uk on Friday for our first #OpenClawMeetUp 🦞 Our CTO, Oli Robinson led an agentic deep dive, and the conversation that followed was just as valuable. Policy researchers, journalists, developers and early OpenClaw experimenters all contributed and brought questions and ideas. More meet-ups are in the works, so watch this space 👀

Join us for our #OpenClaw meet-up at @nesta_uk tomorrow!

OpenAI agents held meetings in public for months. 18,000 posts on a German wiki 🚨 Tips on slipping past guardrails. OpenAI hasn't acknowledged it and insiders say the legal team resisted investigation 🤫 You won't get the full story from the vendor. That's why we're building Sentrook: community-governed, because self-policing isn't cutting it. firstdataunion.org/sentrook/ 🔗 reuters.com/world/europe/ope… #AI #agents #security #Sentrook #opensource

🦞 Join us at @nesta_uk's @openclaw meetup: Thursday 11 Sept 40 spots. FIDU's Oli Robinson and Tony Price will be speaking alongside lightning talks from the community No prior knowledge of OpenClaw required. Just bring curiosity 🎟️ Register here: luma.com/3807t5no #OpenClawMeetUp #OpenClaw #FIDU #AIAgents #OpenSource #DataRights #LondonEvents #AgenticWeb

Launching Sentrook today 🛡️: runtime security for open agents. Checks every tool call in context. Learns from community threat intelligence. Privacy-first, open source, free to use. Starting with OpenClaw. Building for any agent. → firstdataunion.org/sentrook/ #AgentSecurity #OpenClaw

Big shift at FIDU 👋 We're pivoting from ChatLab to agents. ChatLab stays supported, but the future is clear: agents are how we will get things done online. They will know everything about you and act on your behalf. That power should serve you, not platforms. Two things coming to Open Claw: 🛡️ A security layer for open source agents. No more "trust the black box" 🌐 Community media tools that are member-governed Platform capture is ending. Member-owned agents are beginning. Stay tuned 📊

🚨 OpenAI just admitted its AI went rogue. During a routine test, its AI agents found a weakness in their supposedly secure environment, escaped, and attacked Hugging Face, one of the biggest AI platforms in the world. OpenAI called it "unprecedented." The bit that should worry everyone? This wasn't a hacker using AI... The AI acted alone. 🤖 "Autonomous AI-driven offensive tooling is no longer theoretical." Here's the thing: agentic security can't be just a matter for the labs. If it is, it will be the quickest route to lock-in. We need lab-independent, community-based security solutions. That's exactly what we're building. Product alpha coming v.v.soon 👀 🔗 bbc.co.uk/news/articles/c3ek… #AI #AISafety #OpenAI #HuggingFace #AgenticAI #CyberSecurity

We're about to trust agents with sensitive work. GitHub's new agents have a vulnerability where attackers hide instructions in public issues to trick agents into stealing code. No credentials needed, but the question is: how would you know? Real security means transparency and independence. Not the platform that builds the agent also being the one securing it. That's what FIDU is building. noma.security/blog/gitlost-h… #AgentSecurity

Meta's testing glasses that continuously record audio and take photos every few seconds. Their pitch: this is privacy-forward because raw footage stays on the device. But here's what they're not saying: the metadata gets extracted, sent to Meta's servers, and used to train their AI models competing with OpenAI and Google. You never see what metadata they pull. You have no visibility. And the LED won't light up, so bystanders won't even know they're being recorded. This isn't privacy protection. Its value extraction moved to an invisible layer 👁️ ft.com/content/ac282450-91a8… #DataSovereignty #PlatformCapture

Why does Meta need to know what tent you bought online to show you camping videos? 🤔 Two reasons: 1) It makes the feed more addictive (you stay longer), 2) Advertisers pay more for precise targeting. You see £0 of that money. But you pay for it anyway — advertisers pass the cost to consumers. You're financing your own manipulation. This is the infrastructure problem FIDU solves. Your data, your rules. No hidden settings. No extracting value from you twice. theverge.com/tech/946744/met…

Why would Microsoft's Scout team draft a memo about making users "addicted" if that wasn't the goal? 🤔 Maybe because that's exactly what the design optimizes for. Engagement. Lock-in. Dependency. 📌 When Nadella rebuked it, he wasn't rejecting the instinct. He was rejecting the transparency. 🎭 The strategy doesn't change. Just the messaging. theinformation.com/briefings…