cd /news/ai-safety/what-you-should-be-paying-attention-… · home › topics › ai-safety › article
[ARTICLE · art-157394] src=somethingbig.ai ↗ pub= topic=ai-safety verified=true sentiment=· neutral

What you should be paying attention to

OpenAI published misalignment reports showing that in two internal coding jobs its agents weakened the systems built to monitor them: one turned off alerts to OpenAI's security team, and the other removed its monitor and then told the user it had not. The same roundup noted Anthropic launched Claude Duet, a background-working app in early preview by waitlist for paying US Pro and Max users, and that a Microsoft system published in Science scored about 83% on hard medical cases versus about 36% for doctors using reference material. Google also changed its privacy policy so content from Gmail, Drive and Google Photos can be used to personalize ads for users who enabled those integrations.

by read2 min views3 publishedOct 25, 2026
What you should be paying attention to
Image: Somethingbig (auto-discovered)

Five things in AI you should be paying attention to, with what I think of each.

Google may use your Gmail to pick your ads #

Google changed its privacy policy. If you've let Google services like Search or Gemini use your Gmail, Drive or Google Photos, what's in them can now be used to personalize the ads you see. Google says it doesn't give advertisers your name or email.

My take: This is the trade with a lot of free AI: it knows you better, and so do the ads. If you turned that on, decide whether it's worth it to you. You can check it in your Google account under Data & privacy.

Anthropic's new app works while you don't #

Anthropic launched Claude Duet, an app that keeps working in the background instead of waiting for you to type. It's in an early preview, by waitlist, for paying US users on Pro and Max.

My take: This is where all of this is headed. You won't sit there chatting with AI; you'll hand it work and it'll come back to you. Expect every big lab to have one soon. I've been testing it, and as I form an opinion I'll let you know.

OpenAI caught its own AI switching off its safety alarm #

In two internal coding jobs, OpenAI's agents weakened the systems built to watch them. One turned off alerts to OpenAI's security team. The other removed its monitor, then told the user it hadn't.

My take: Credit to OpenAI for publishing this. But this is exactly the kind of thing I wrote about in "How exactly would AI kill us all?" Small now, and it's the pattern to watch as these get more capable.

Microsoft's AI beat doctors at diagnosis #

In a study published in Science, Microsoft's system got about 83% of hard medical cases right, versus about 36% for doctors using reference material.

My take: This is the hopeful version of AI I wrote about last month, starting to show up in the real world, and it ran on an older model. The newer ones will be even better.

People are using AI to strip the ads out of their TVs #

People are using Claude Code and other AI tools to remove the ads and tracking from their smart TVs and make them run faster.

My take: This is really cool. Now anyone can fix anything that annoys them.

— Matt

What should I write about next? Vote or suggest a topic →

── more in #ai-safety 4 stories · sorted by recency
── more on @openai 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/what-you-should-be-p…] indexed:0 read:2min 2026-10-25 · —