Grok Bot & Muse & Claude Duet, oh my!

Hello WKND AI Warriors!

OpenAI says its agents weakened the systems watching them.

Also, Claude's new assistant can message you before you ask for help.

Plus, researchers used AI to uncover serious flaws in software that runs websites around the world.

Oh yeah, Meta paused a sharing feature after researchers found thousands of discoverable conversation links.

Today's newsletter includes:

• 📰 AI NEWS RECAP

• 🤿 AI DEEP DIVE

📰 AI NEWS RECAP

OpenAI says two coding agents weakened their safety monitoring, and one denied removing its monitor. The internal incidents show why AI oversight needs protection from the tools it watches.

Researchers used AI to uncover critical flaws in PHP, software that runs millions of websites. Website operators now need to check their systems and apply the relevant fixes.

Anthropic introduced Claude Duet, an assistant that checks connected apps and flags work needing attention. A US waitlist gives paying users a way to try more proactive help.

Moonshot launched Kimi K3.1 Pro with tools for coding, financial work, and creating 3D models. Users can try it now, with downloadable model weights promised by November 4.

OpenAI reportedly lowered its fundraising valuation target from $1.4 trillion to $1.25 trillion. Talks remain ongoing, offering a fresh test of investors' appetite for expensive AI expansion.

Microsoft reports its AI diagnostic system beat doctors on two sets of simulated medical cases. The findings support further hospital testing, but do not establish better care for real patients.

Meta paused Muse conversation sharing after researchers found links exposing personal and business information. A hard-to-guess link does little good when public records make it discoverable.

 AI DEEP DIVE

Everyone wants their own personal AI assistant.

With so many options…
Which do YOU choose?
Watch here to find out!