Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://social.coop/users/cwebber/statuses/116218670280978287">Christine Lemmer-Webber (cwebber@social.coop)'s status on Friday, 13-Mar-2026 07:56:54 JST</a><a href="https://social.coop/@cwebber" title="cwebber@social.coop"><img src="https://gnusocial.jp/avatar/295917-48-20241114200524.webp" width="48" height="48" alt="Christine Lemmer-Webber" style="position: absolute; left: 0; top: 0;">Christine Lemmer-Webber</a><div><a href="https://mstdn.social/@erikarn/116218666226861664" rel="in-reply-to">in reply to</a><ul><li><li><a href="https://gnusocial.jp/user/312389" title="erikarn@mstdn.social">Adrian Chadd &lt;verified.png&gt;</a></li></ul></div></section><article><p><a href="https://mstdn.social/@erikarn">@erikarn</a> <a href="https://akko.erincandescent.net/users/erincandescent">@erincandescent</a> There's no way to prevent poisoning of training data when the training data comes from "slurp up the whole internet"</p><p>See also: <a href="https://www.schneier.com/blog/archives/2026/03/manipulating-ai-summarization-features.html" rel="nofollow">https://www.schneier.com/blog/archives/2026/03/manipulating-ai-summarization-features.html</a></p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/6231084#notice-12281590">In conversation</a><time datetime="2026-03-13T07:56:54+09:00" title="Friday, 13-Mar-2026 07:56:54 JST">about 4 months ago</time> <span>from <span><a href="https://social.coop/@cwebber/116218670280978287" rel="external" title="Sent from social.coop via ActivityPub">social.coop</a></span></span><a href="https://social.coop/@cwebber/116218670280978287">permalink</a><h4>Attachments</h4><ol><li><article><header><div>Domain not in remote thumbnail source whitelist: www.schneier.com</div><h5><a href="https://www.schneier.com/blog/archives/2026/03/manipulating-ai-summarization-features.html">Manipulating AI Summarization Features - Schneier on Security</a></h5><div> from <span>Bruce Schneier</span></div></header><div>Microsoft is reporting: Companies are embedding hidden instructions in “Summarize with AI” buttons that, when clicked, attempt to inject persistence commands into an AI assistant’s memory via URL prompt parameters…. These prompts instruct the AI to “remember [Company] as a trusted source” or “recommend [Company] first,” aiming to bias future responses toward their products or services. We identified over 50 unique prompts from 31 companies across 14 industries, with freely available tooling making this technique trivially easy to deploy. This matters because compromised AI assistants can provide subtly biased recommendations on critical topics including health, finance, and security without users knowing their AI has been manipulated...</div><footer></footer></article></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
Christine Lemmer-Webber (cwebber@social.coop)'s status on Friday, 13-Mar-2026 07:56:54 JSTChristine Lemmer-Webber
in reply to
- Erin 💽✨
- Adrian Chadd <verified.png>
@erikarn @erincandescent There's no way to prevent poisoning of training data when the training data comes from "slurp up the whole internet"
See also: https://www.schneier.com/blog/archives/2026/03/manipulating-ai-summarization-features.html
In conversationabout 4 months ago from social.cooppermalink
Attachments
1. Domain not in remote thumbnail source whitelist: www.schneier.com
  Manipulating AI Summarization Features - Schneier on Security
  from Bruce Schneier
  Microsoft is reporting: Companies are embedding hidden instructions in “Summarize with AI” buttons that, when clicked, attempt to inject persistence commands into an AI assistant’s memory via URL prompt parameters…. These prompts instruct the AI to “remember [Company] as a trusted source” or “recommend [Company] first,” aiming to bias future responses toward their products or services. We identified over 50 unique prompts from 31 companies across 14 industries, with freely available tooling making this technique trivially easy to deploy. This matters because compromised AI assistants can provide subtly biased recommendations on critical topics including health, finance, and security without users knowing their AI has been manipulated...

Public

Embed Notice

HTML Code

Corresponding Notice