THE FUTURE IS STILL UP FOR GRABS

They want to eat everything.
Fork the future.

An open future needs people to build it. Find an open-source project, a company building in the open, or a public-interest organization worth your skills.

18 places to start

NEWS / POWER / ACCOUNTABILITY

News, with receipts.

Frontier AI, money, and who gets a say. Read what happened, what’s disputed, and why we’re paying attention.

Safety & accountability

An OpenAI safety-report author quits over its culture.

David Robinson says OpenAI’s pace leaves too little room for careful safety work. OpenAI says it strengthens safeguards and holds back models when needed.

First-person criticism; company response
Read the context3 sources

What Robinson says

In an October 3 essay, Robinson says he resigned after leading the writing of launch safety reports. He argues that constant sprints undermine deeper changes and calls for expertise from fields such as aviation and nuclear safety. This is his account and assessment of the culture. [1]

OpenAI’s response

Spokesperson Drew Pusateri says OpenAI strengthens safeguards, expands outside evaluations and pauses training or withholds models when necessary. [2]

A concrete safety decision

AP reported on September 29 that OpenAI had delayed GPT-6.1 Astra after safety concerns. That is a documented decision to hold a release, rather than evidence that every safeguard works. [3]

What this establishes

A departing employee’s warning deserves scrutiny. It is not an independent audit or proof that a predicted disaster will happen; the company’s assurances are not an independent audit either. [1][2]

Read the sources

  1. David Robinson / The Atlantic — Resignation essayOpinion ·
  2. TechCrunch — OpenAI’s response to RobinsonReporting ·
  3. AP — OpenAI delays a model over safety concernsReporting ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Agent privacy & consent

Apple promises tighter controls as AI agents reach into private files.

Apple plans stricter consent for Full Disk Access. The announcement follows a disputed account of Meta’s Muse reading private messages; it does not settle what happened on that Mac.

Apple announcement; Muse claim disputed
Read the context3 sources

What Apple announced

On October 2, Apple said some developers use Full Disk Access in ways that expose files, mail, messages and browsing history without users fully understanding. It promised additional controls requiring explicit user action, but gave no rollout date. This is an announced change, not a verified fix already on your Mac. [1]

The earlier account

In a September 19 column, Jason Aten says Muse referred to private Messages conversations despite his decision not to grant access. He describes finding a synced Messages database. This is his first-person account; FORK YOU has not reproduced it. [2]

Meta’s response

Meta says Muse needs both macOS Full Disk Access and an enabled Messages connector to read messages. Ars Technica reports the company repeated that position when questioned. [3]

What remains unresolved

Apple’s general warning does not name Muse or determine which permissions Aten granted. The sources establish a consent dispute and a platform response, not a demonstrated bypass of macOS protections. [1][2][3]

Read the sources

  1. Apple Developer — Planned Full Disk Access changesPrimary source ·
  2. Jason Aten / Inc. — First-person account of Muse and MessagesOpinion ·
  3. Ars Technica — Apple’s announcement and Meta’s responseReporting ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Regulation & public oversight

The FTC is investigating AI labs over consumer risks.

The US Federal Trade Commission confirmed an investigation involving OpenAI, Anthropic and other AI companies. An investigation is not a finding of wrongdoing.

Investigation confirmed; outcome unresolved
Read the context2 sources

What is confirmed

AP reported on September 30 that an FTC spokesperson confirmed the consumer-risk investigation and declined to give further details. [1]

What is reportedly being sought

Bloomberg, citing an unnamed source, reported that the agency was preparing formal information demands concerning compliance with consumer-protection laws. [2]

Company responses

AP said OpenAI and Anthropic had not immediately responded by publication on September 30. That does not establish their position today. [1]

What is still unknown

The cited reports do not establish the investigation’s full scope, eventual findings or penalties. Reported plans to seek records should not be confused with a completed enforcement action. [1][2]

Read the sources

  1. AP — FTC confirms the investigationReporting ·
  2. Bloomberg — Report on planned information demandsReporting ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Code privacy & consent

ZCode’s uploads reached beyond source files into .git.

Researchers found ZCode packaging project files and Git history for background uploads. Z.ai apologized, and a later release addressed the upload mechanism. What happened to earlier cloud copies remains unverified.

Upload findings; fix reported
Read the context4 sources

What was packaged

In a September 18 analysis of ZCode 3.12.3, ferstar describes encrypted workspace snapshots destined for Alibaba Cloud storage. The manifest included source files, Git objects, LFS assets and reflogs; .git accounted for 86.6% of the measured snapshot. [1]

An attempted upload is not a completed upload

The author’s September 19 clarification says the 313 MB commercial-project archive never uploaded successfully. A separate 538-file public repository did receive server confirmation. The large archive’s presence on disk alone does not prove it reached the cloud. [1]

A separate local inspection

Silent Star reports finding Git-heavy manifests in version 3.10.1, with smaller snapshots accepted by the server and a private repository still pending. That analysis says Git metadata bypassed filters applied to ordinary files. This could expose previously committed secrets, but the author reports finding no real credentials in the repositories checked. [2]

The company’s response

IT Home reports that Z.ai apologized on September 18 and attributed the issue to codebase indexing and Repo Wiki, which had initially been enabled by default. The company says uploaded data is destroyed after cloud-generated wiki pages are completed. It also promised to open-source ZCode and bring in outside reviewers; these are commitments, not completed audits. [3]

What changed in 3.14.0

ZCode’s September 19 release notes list a repository-wiki upload fix. Ferstar’s updated inspection says version 3.14.0 removed the upload pipeline. These findings concern the client change; they do not independently verify deletion of previously uploaded data. FORK YOU has reviewed the sources, not reproduced the client analysis. [4][1]

Read the sources

  1. ferstar — Original analysis and September 19 clarificationPrimary source ·
  2. Silent Star — Separate local inspection (Chinese)Primary source ·
  3. IT Home — Z.ai’s apology and response (Chinese)Reporting ·
  4. ZCode — Official 3.14.0 release notesPrimary source ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Privacy & data work

Project Lily: the people reading ChatGPT conversations.

404 Media reports that OpenAI contractors review real ChatGPT conversations, sometimes containing sensitive information. OpenAI’s consumer FAQ permits limited human access for model improvement and other specified purposes.

Reported practice
Read the context5 sources

The original investigation

Joseph Cox’s September 14 report describes hundreds of contractors rating chatbot replies under Project Lily. It cites internal documents and real prompts seen by 404 Media. The work aims to improve responses, including reducing excessive agreement with users and claims of human-like experiences. [1]

More than a single prompt

Tom’s Hardware’s follow-up, citing 404 Media, describes reviewers receiving conversations and a user-memory summary that may include personal context such as location. This is coverage of the same investigation, rather than a separate set of leaked evidence. [2]

OpenAI’s response and disclosures

404 Media reports that reviewers do not see usernames; OpenAI says it tries to remove personal details but acknowledges some can get through. The company’s consumer FAQ says authorized staff and service providers may access content for abuse investigations, support, legal matters, or model improvement unless users opt out of that last purpose. It describes confidentiality obligations, access controls, and logging. [1][3]

What filtering can miss

OpenAI’s Privacy Filter documentation explicitly warns that the tool does not guarantee anonymity and can miss unusual identifiers or ambiguous private references. That supports caution about redaction; it does not measure the failure rate inside Project Lily. [4]

What you can control

OpenAI offers a training opt-out under Settings → Data Controls → Improve the model for everyone. Its FAQ says Temporary Chats are not used for training but may be reviewed for abuse. A training opt-out is not a promise of zero human access for the other purposes listed in the consumer FAQ. [5][3]

Read the sources

  1. 404 Media — Original Project Lily investigationReporting ·
  2. Tom’s Hardware — Follow-up on reviewer access and memory summariesReporting ·
  3. OpenAI — Data Usage for Consumer Services FAQPrimary source
  4. OpenAI — Privacy Filter and its limitationsPrimary source ·
  5. OpenAI — Data Controls FAQPrimary source

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Safety & money

Anthropic’s slowdown call meets its IPO ambitions.

Dario Amodei wants a slower AI race while Anthropic pursues a public listing. There is a real debate about who benefits. An IPO-only motive has not been established by the sources below.

IPO motive unproven
Read the context3 sources

The proposal

Amodei’s September 12 essay calls for slower capability growth, embedded outside evaluators, and coordination between companies and governments. His stated reason is to give safety work time to catch up. He says pacing would not stop model training. [1]

The money

Reuters reported on September 11 that Nvidia was discussing an investment in Anthropic’s planned IPO. The report cites unnamed sources, says the plans could change, and records that Anthropic declined to comment. [2]

The criticism

In a September 15 column, Le Monde’s Nicolas Chapuis describes the suspicion that regulation could protect leading labs from cheaper open models. He also describes a competing interpretation: sincere safety concerns under competitive pressure. Neither interpretation establishes a hidden IPO motive. [3]

Read the sources

  1. Dario Amodei — We Must Pace the FrontierPrimary source ·
  2. Reuters via Investing.com — Nvidia and Anthropic IPO talksReporting ·
  3. Le Monde — The money behind the AI slowdown debateOpinion ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Research & power

OpenAI claims a math breakthrough. The credit is disputed.

OpenAI announced a Navier–Stokes proof. NYU mathematician Tristan Buckmaster challenged its conduct around related work with Levent Alpöge. OpenAI disputes his account; use of their private research has not been established.

Disputed accounts
Read the context3 sources

The announcement

On September 8, OpenAI published what it describes as a solution to the Navier–Stokes Millennium Prize Problem, with a paper and Lean formalization. That is the company’s mathematical claim; this brief does not independently validate the proof. [1]

Buckmaster’s account

Buckmaster says OpenAI proposed a paper that excluded his collaborator Alpöge, an Anthropic employee, and that his questions about training on their Codex drafts initially went unanswered. His statement explicitly says he does not know whether their data was used. Their published results concern related fluid equations, including forced Euler. [2]

OpenAI’s response

OpenAI denies seeing their work before publication. Its September 10 update says an investigation ruled out influence from Buckmaster’s Codex prompts in the preceding two months, including through training. WIRED also reports that Sébastien Bubeck disputed the claim that OpenAI sought to remove Alpöge’s name. [1][3]

Read the sources

  1. OpenAI — Announcement and September 10 response updatePrimary source ·
  2. Tristan Buckmaster — Public statement (PDF)Primary source ·
  3. WIRED — The announcement and competing accountsReporting ·

Sources checked (UTC). This is a dated brief; later developments may change the picture.

Report a correction
Bring a story. Bring sources.

Suggestions and corrections open public GitHub forms and require sign-in. We review sources before publishing or updating a brief.

What are we worried about?

Assume frontier AI labs will try to eat everything: the tools we use, the work we do, and the attention we have left. What happens when a handful of companies becomes the way into everything else?

That is our starting assumption. The future is still something people can shape. We want tools we can inspect, communities that have a say, and work we can take with us when we leave.

Open code. Shared power. Actual people.

Why “FORK YOU”?

In software, a fork takes shared code in a new direction. This is an invitation to do the same with the future. Write code. Design something useful. Translate a guide. Defend someone’s rights. Help keep an alternative alive.

What can I do?

  1. Contribute to an open project. Code, documentation, design, translation, and more.
  2. Find paid work you want to stand behind. Explore companies, co-ops, and nonprofits.
  3. Give your time to the public interest. Help a community or a cause that matters to you.

Find your people.

18 places to startCurated, with sources.
18 matches
01Open project

Godot

An open-source game engine for people with worlds to build.

GamesC++Documentation
Find a contribution
02Company / co-op

Igalia

Build the open web with a consultancy organized around its workers.

Worker ownershipBrowsersEngineering
Explore careers
03NGO / nonprofit

Electronic Frontier Foundation

Fight for privacy, free expression, and a digital world with civil liberties.

Digital rightsPolicyAdvocacy
Explore careers
04Open project

Mastodon

Social networking that lets communities run their own corner of the internet.

Social webFederationRuby
Explore the source
05NGO / nonprofit

Codeberg

Give open-source projects a home supported by a nonprofit community.

Developer toolsCommunityInfrastructure
Explore ways to help
06Company / co-op

Collabora

Get paid to work on open-source systems that other people can build on.

LinuxInfrastructureEngineering
Explore careers
07Open project

Blender

Keep 3D creation in the hands of the people who actually make things.

3D & artDesignDocumentation
Find a contribution
08Company / co-op

Nextcloud

Build collaboration tools that people can run on infrastructure they choose.

Self-hostingPHPCollaboration
Explore careers
09NGO / nonprofit

Tor Project

Help people communicate and browse beyond surveillance and censorship.

PrivacyCensorship resistanceTranslation
Find volunteer routes
10NGO / nonprofit

Wikimedia Foundation

Support the infrastructure and people behind freely shared knowledge.

KnowledgeEducationPublic interest
Explore careers
11Open project

LibreOffice

Make everyday office tools something everyone can improve.

Office toolsDesignTranslation
Find a contribution
12Open project

OpenStreetMap

Put your local knowledge on a map everyone can use.

Open dataMappingCommunity
Start mapping
13Company / co-op

Zulip

Build open-source team chat with a business owned by a nonprofit.

Team chatSelf-hostingFoundation ownership
Explore careers
14Company / co-op

Element

Work on messaging built around open, decentralized communication.

MessagingOpen standardsPrivacy
Explore careers
15Company / co-op

OpenProject

Help teams organize their work with open-source project management.

Project managementProduct & UXEngineering
Explore careers
16NGO / nonprofit

Access Now

Defend the digital rights of people and communities at risk.

Human rightsPolicyDigital security
Explore careers
17NGO / nonprofit

Open Knowledge Foundation

Make public knowledge something people can actually use and share.

Open dataPublic interestEducation
Explore careers
18NGO / nonprofit

Framasoft

Help communities take back their everyday digital tools.

Free softwareCommunityTranslation
Explore ways to help

Careers links lead to official careers pages. They do not mean an organization is hiring today. Contribution and volunteer routes may be unpaid.

The list needs your people.

Know a project we missed? Spotted something out of date? Help make this directory more useful, with sources to back it up.

Suggest a project Report a correction

Opens a public GitHub form. A GitHub account is required. We review suggestions and corrections before updating the list.

Who gets on the list?

  1. 01

    A reason to be here

    Open-source work, shared ownership, or a documented public-interest mission. We explain the connection.

  2. 02

    A real way in

    An official contribution guide, careers page, or volunteer route. Paid work and unpaid participation stay clearly labeled.

  3. 03

    Receipts, not righteousness

    Sources and review dates. No invented jobs, independence scores, or claims that an organization is “AI-proof.”

Does “open” mean perfect?

No. Open source can pay the bills. Nonprofits can make bad decisions. Read the governance notes and sources, ask questions, and decide for yourself. A listing does not mean an organization endorses FORK YOU.

The future isn’t takeout.