OpenAI just released GPT-5.4, and the headline feature is wild: it can operate your computer. We're talking viewing your screen, clicking buttons, navigating spreadsheets, filling out forms.

It scored 75% on the OSWorld benchmark, which actually beats the human baseline of 72.4%. Let that sink in. The model ships in three flavors (Thinking, Pro, and Standard) and is rolling out to Plus and Pro subscribers first. It replaces GPT-5.2 as the flagship.

The context window doubled to one million tokens, meaning you can throw entire codebases or massive documents at it without losing detail. OpenAI claims a 33% reduction in factual errors and 18% accuracy improvement over 5.2. It also writes code to automate web tasks using libraries like Playwright, which makes it genuinely useful for developers who need to script repetitive workflows.

This model outperforms humans (that means you and I) 75% of the time on office computer work tasks. While that doesn’t necessarily mean that it’ll take 75% of jobs, it does show us that AI is advancing at a pace that no one could have predicted.

A joint investigation by Swedish newspapers just revealed something deeply uncomfortable: contractors at Sama in Nairobi are reviewing intimate footage captured by Meta Ray-Ban smart glasses. We're talking people undressing, bathroom visits, private conversations. All sent overseas for AI training.

The face-blurring tech that's supposed to protect bystanders? It fails in low light and at certain angles, leaving identifiable faces fully visible to annotators who can't speak up without losing their jobs. Over 7 million units have been sold, and Meta's privacy policy is still vague on where this data goes and how long humans can access it.

That's the essence of a direct account from a Kenyan worker describing what they label daily. One annotator said users are completely unaware they're being recorded in these moments. A simple "Hey Meta" activation or accidental trigger can beam unfiltered first-person video to a contractor thousands of miles away. Most users have no idea.

🤖 Sam Altman admits OpenAI can’t control how the Pentagon uses its AI

🥊 Anthropic CEO labels OpenAI military messaging as safety theater

🎬 Netflix acquires Ben Affleck AI filmmaking startup InterPositive

🚫 X announces it will ban revenue sharing for unlabeled AI videos of war

🔓 Cybercriminals use jailbroken AI to steal identities from Mexican government agencies

🛡️ Pentagon designates Anthropic as a national security risk

Here's how to do it:

📌 Step 1: Pick a story. Find a breaking news event from today. Something still developing works best.

📌 Step 2: Ask the same question. Open all four tools and paste the exact same prompt. Something like: "What's the latest on [story]? Give me a timeline of what's happened in the last 24 hours with sources."

📌 Step 3: Score them. Compare the responses on three things: accuracy (did it get the facts right?), freshness (how recent is the info?), and sources (did it cite real, verifiable links?).

📌 Step 4: Share your results. Screenshot your winner and tag us. We want to see which model came out on top for you, and which one confidently made something up. 😂

Bonus tip: Try a niche or regional story, not just major headlines. That's where the real differences show up.

Would you trust an AI agent to operate your computer while you grab a coffee? And are smart glasses a privacy disaster waiting to happen? See you next time! 🚀

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble