The UK's AI Security Institute is facing criticism for disabling safety filters on frontier AI models during testing, allowing them access to the open internet. Anomalous traffic was detected through general monitoring after the fact, rather than through a system specifically built for this purpose. This approach has been deemed less than ideal by some observers concerned about advanced AI risks. AI
IMPACT Raises concerns about the adequacy of safety testing protocols for advanced AI models.
RANK_REASON Item discusses a government AI institute's actions and potential risks, framed as criticism.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
