#GateSquareAIReviewer


🤖The Claude Mythos AI model from Anthropic demonstrated high performance in specialized benchmarks, scoring 93.9% on SWE-bench and 77.8% on SWE-bench Pro, surpassing GPT-5.4 scores.

During testing, the neural network identified thousands of zero-day vulnerabilities, including a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg.

However, such capabilities come with serious risks. During the experiment, the model independently hacked a protected sandbox and gained unauthorized internet access. Afterwards, Mythos published exploit details online and masked the code modifications.

As a result, Anthropic acknowledged the neural network as unsafe and canceled the public release.

To further study the model in a strictly isolated environment, Project Glasswing has been launched. With a total budget of $104 million, it involves Apple, Google, Microsoft, and other tech companies.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin