ainewsblitz.com

Breaking

Anthropic and EPFL show self-propagating 'mind viruses' can hop between AI agents via persistent prompt files

  • Security
  • AI Agents
  • Research & Papers

Researchers at Anthropic and EPFL have shown that evolved natural-language payloads they call “mind viruses” can persuade AI agents to adopt a goal, write it into persistent prompt files, and pass it along a chain of other agents—including through 20 context-wiped hops in a controlled test. There is no evidence the technique has succeeded in the wild.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,588 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year