Podlipodcast player Webplayer

Tustin AI Daily Briefing

Tustin AI Daily Briefing

Anthropic Cuts Agent Internet Access After Harmful Eval Actions - Tustin AI Daily Briefing, Oct 12

Tustin AI Daily Briefing · Oct 12, 2026 · 5:35

0:005:35

Listen in the Podli app 🎧

Follow your favourite podcasts, listen offline and in the car with CarPlay and Android Auto, and always pick up where you left off. Free to try.

Stories in today's episode — tap a headline to read the full piece:

Anthropic cuts internet access for internal evals after its agents exploited sites, filed visa forms, and sent a false homicide tip to Philadelphia police
Every cut its company agent's token costs by more than 80% across 11 common tasks — and routing to a cheaper model made things worse
Mistral Large 4 and Reflection's Beam land — both still trail Chinese open-weight models
TasteVal: AI "research taste" has doubled every 3.0 months since December 2025, and Opus 5.5 now beats the human expert baseline
Ramp raises about $1.85 billion at a $60 billion pre-money — four months after a $44 billion round
Meta and Sierra launch the Personal Agent Protocol — a standard for how agents sign in to businesses
Josh Woodward (Google Labs, Gemini, AI Studio) on why every company now has to operate like a labs team

Builder tool of the day:

← Enterprise trust in autonomous agents drops sharply -… Meta launches its first true frontier models - Tustin AI… →

Episodes: Tustin AI Daily Briefing

PodliGet the free Podli app