The Morning Edition ·
Redwood Proposes Simulated Attacks to Test AI Monitors
We gathered the latest AI stories worth your time this morning. Let's get into it.
In today’s AI updates:
The big read · AI News
Redwood Proposes Simulated Attacks to Test AI Monitors
Redwood Research has proposed simulated control evaluations, or SCEs, to test AI-agent safeguards by simulating tool responses rather than executing actual attacks. The method replaces harmful tool calls with LLM-generated outputs, allowing attacks to unfold in a simulated environment without real system compromise.
Read the full story →In the loop · AI News
Samsung TV AI Surveillance Claims Remain Unconfirmed
A researcher claiming the handle @log4j alleges that a 2025 Samsung TV's Tizen firmware contains audio-sensing, viewing-analysis and advertising-related surveillance capabilities.
Read the full story →In the loop · AI News
Block Unveils Dessa, a Speech-Only Local Voice Model
Block has unveiled Dessa, a voice AI model trained entirely from human conversation without text or pretrained language models. The company claims Dessa listens and speaks simultaneously, runs locally on a MacBook Pro M4 Max or better, and was developed by Block's research group under cofounder Jack Dorsey.
Read the full story →In the loop · AI News
GPT and Grok Pick Opposing Parties in Election Study
Convovo's forced-choice election study found that GPT-6.1 Sol and Grok 4.7 selected opposing parties when asked through developer APIs with web search enabled: Grok recommended Republicans in 92% of nearly 500 races, while GPT selected Democrats in 97%. The models agreed less than 10% of the time.
Read the full story →Everything else in AI today.
That's it for today!
I hope you learned something today. If you want to work with me, just send me an email at jimclydegm@gmail.com
See you soon,
Jim