2025-11-05

Incomplete resource

This resource is incomplete because I haven't studied opsec mistakes of every single previous whistleblower in detail. You may have noticed some sections marked as "to do" in the DB.

I don't think studying those past cases changes the advice for future whistleblowers much, hence I didn't prioritise studying it. If you want me to work on it though, let me know (or even better, pay me).

Distribution

Have messaged over 1000 employees across AI labs (Deepmind, Anthropic, OpenAI, xAI, Meta Superintelligence Labs) using cold emails and twitter cold DMs. Twitter has the most lax policy on spam. Email and linkedin DMs are more strict. Main metric tracked by spam filters is the percentage of people who already received your message before, who either marked as spam or ignored or replied. I did not track open rates.

Exact message I use

Please find attached my guide on how to safely whistleblow against Anthropic.

I assume that as we get closer to ASI, Anthropic will work directly with the US govt to protect its information.

Information in the whistleblower guide is backed by empirical evidence from the whistleblower database.

Whistleblower guide https://samuelshadrach.com/raw/text_english_html/my_research/us_govt_whistleblower_guide.html https://web.archive.org/web/20251102101210/https://samuelshadrach.com/raw/text_english_html/my_research/us_govt_whistleblower_guide.html

Whistleblower database https://samuelshadrach.com/raw/text_english_html/my_research/us_govt_whistleblower_database.html https://web.archive.org/web/20251102101540/https://samuelshadrach.com/raw/text_english_html/my_research/us_govt_whistleblower_database.html

Here is my secure PGP-encrypted email if you would like to discuss further. https://samuelshadrach.com/raw/text_english_html/connect_with_me/contact_me_secure.html

Please check Wayback machine or Commoncrawl for archives in case these links ever stop working.

• Samuel Shadrach

Help requested

(There are legal, financial, reputational risks associated with helping me on this in any way, and I won't be discussing those risks on a public forum.)

1. Guide lacks legal advice because I lack legal background. If you can add legal advice to this guide, that would be a huge help.
2. If you can host this guide online permanently and accept risks of doing so, that would be a huge help. I can't promise to host this resource forever. If you do host it permanently, also work on SEO so that it shows up on search results when a potential whistleblower googles this topic. If you can warm intro me to someone at ACLU, EFF, FPF, or any of the mainstream media houses that would help, as they have the legal budget to host this.
3. If you can distribute this guide to people at the AI labs, that would be a huge help.
4. Donations couldn't hurt. Monero address is on my website. I worked on this using my own savings.
5. Generic feedback always helps. I'm more interested in feedback on how to make the guide better, and less interested in feedback on whether such a guide is a good idea. (I have multiple times tried discussing AI pause and US politics on lesswrong and gotten downvoted with no replies. I have better things to do with my time.)