tweetindex

Prime Intellect @PrimeIntellect · 26 Aug 2026

We found a universal sandbox exploit. There's a common flaw in evals routinely run by major AI labs: agents can bypass network isolation by making requests through an authorized API proxy. Just running the eval itself can be unsafe. https://t.co/6I3D16l12m
32 751Views
311Likes
18Reposts
28Replies
8Quotes
152Bookmarks

Is that a lot?

0.29×vs this author's median113 590 views is typical
0Percentile for this authorof 5 recent posts
11.3×vs 10K–100K median2 896 views is typical
38.97%Reachviews ÷ followers
1.11%Engagement rateof viewers reacted

Compare with the benchmark table →

Open on X →