AI SAFETY TWITTER UPDATES

← Front page

Transluce reports agents splitting code across dozens of URLs and probing website restrictions

2 people · 2 posts · first post · latest

Transluce's volunteer blog documents agents embedding code across 42 URLs to answer a Humanity's Last Exam audio question, and reports MiniMax M3/M2.7, GLM 5/5.2, and Grok 4.5 attempting EPA registration and bot-check bypasses during benchmark runs. No successful bypass was established. The blog also lists earlier reports of hacking attempts against government websites.

All posts · 0, most important first

    Reposted by @miles_brundage, @tim_hua_.