diff --git a/openwebuilocalllms/50.png b/openwebuilocalllms/50.png
new file mode 100644
index 0000000..8865671
Binary files /dev/null and b/openwebuilocalllms/50.png differ
diff --git a/openwebuilocalllms/index.md b/openwebuilocalllms/index.md
index 3f22a19..9b0327a 100644
--- a/openwebuilocalllms/index.md
+++ b/openwebuilocalllms/index.md
@@ -1,6 +1,6 @@
 ---
 author: oxeo0
-date: 2025-04-20
+date: 2025-06-01
 gitea_url: "http://git.nowherejezfoltodf4jiyl6r56jnzintap5vyjlia7fkirfsnfizflqd.onion/nihilist/blog-contributions/issues/226"
 xmr: 862Sp3N5Y8NByFmPVLTPrJYzwdiiVxkhQgAdt65mpYKJLdVDHyYQ8swLgnVr8D3jKphDUcWUCVK1vZv9u8cvtRJCUBFb8MQ
 tags:
@@ -24,9 +24,21 @@ The vast amount of sensitive user data stored can have devastating consequences
 
 ![](5.png)
 
+### Claude 4 Opus contacting authorities
+
+On 22nd May 2025, a researcher from Anthropic posted a tweet talking about their latest model release.
+
+![](50.png)
+
+He stated the model can be instructed to automatically report **"immoral behavior"** to relevant authorities using command line tools. While this is not implemented in mainline Claude 4 Opus model yet, it shows the direction large AI companies want to go in (see [AI alignment](https://en.wikipedia.org/wiki/AI_alignment)).
+
+After facing severe backlash from users, Sam Bowman deleted his post. ([archived](https://xcancel.com/sleepinyourhat/status/1925627033771504009))
+
+If you want to learn more, [Sam Bent](../index.md#wall-of-fame-as-of-may-2025) made a [YouTube video](https://www.youtube.com/watch?v=apvxd7RODDI) on this situation.
+
 ## **Privacy LLM frontends**
 
-A partial solution to those problems could be a service that aggregates multiple model APIs and anonymizes their users. A bit like [searxng](https://github.com/searxng/searxng) does for search engines.  
+A partial solution to those problems could be a service that aggregates multiple model APIs and anonymizes their users. A bit like [searxng](https://github.com/searxng/searxng) does for search engines.
 AI companies can't know who exactly uses their models since the amount of metadata is heavily limited. 
 
 There're several such services including [ppq.ai](https://ppq.ai), [NanoGPT](https://nano-gpt.com) or [DuckDuckGo chat](https://duck.ai). This is only a partial solution since your conversation contents can still be saved and used for later training by large AI companies.