Not even AI companies know what to do when it goes wrong

The "t" in t-shirt stands for the shape it makes when laid flat. A capital T.

You have been wearing the answer your whole life and never once looked down.

What’s in store:

OpenAI wins a very weird leaderboard.
LinkedIn’s AI slop button is getting absolutely hammered.
Mindstream Picks: Tesla has reportedly killed off its Solar Roof.
Learn how to nail AI graphics every single time!

Read Time: 5 minutes

AI SAFETY

"Just unplug it” is not an AI safety strategy

A new study suggests some of the world’s biggest AI companies have shared very little about what they would do if one of their models started acting outside human control.

Guidelight AI Standards reviewed public safety plans from OpenAI, Anthropic, Google, Meta and xAI, looking at things like monitoring, shutdown procedures and what happens when a model keeps breaking safeguards.

OpenAI scored highest, while Anthropic and Meta scored lowest.

However, the study only looked at public information, so companies may have stronger internal plans they haven’t shared.

It has also been suggested that publishing such a plan could be a legal own-goal if it doesn’t work, and could be the basis of a deceptive marketing lawsuit.

In brief:

OpenAI scored 3/5 and ranked highest.
Anthropic and Meta scored lowest for public containment plans.
The main gap: labs talk more about testing models before release than what happens if one misbehaves afterwards.

Have we considered… turning it off?

The issue is becoming more important as AI agents get more access to company systems and can take more actions on their own.

Regulators are also starting to step in, with new rules in California and New York requiring major AI developers to explain how they would handle serious safety incidents.

I’m sure they have a more serious plan internally, right? Right? - MV

How important is a public “rogue AI” shutdown plan?

Non-negotiable

Internal plans are enough

Vote for live results + see results and opinions from yesterday at the bottom of the email.

AI DESIGN

Stop Getting 6-Fingered Monsters From Your AI Image Prompts

AI image tools are powerful — and absolutely feral if you prompt them wrong. Six fingers, melted text, ears that belong on a different species. The chaos is real. That's why Rory Flynn, AI design educator and prompt strategist, built a set of visual prompting workflows that actually produce polished, professional-grade output.

Get instant access to:

Rory Flynn's Workflows step-by-step flows for images, video, and photoshoots
Pro-Grade Output prompts engineered for clean, publish-ready visuals
No Design Degree needed — just follow the flow chart
Faster Than Agencies results in hours, not days

Go from chaotic outputs to consistent, studio-quality AI visuals.

Grab The Workflows →

Number Crunch

A pool is 50m long. You swim 3km. How many laps (there and back) is that?

Find the answer at the bottom of the email!

That’s Weird

These robots may have speed, but we’re still waiting on their dignity

A Chinese humanoid robot called Tiangong Ultra ran 100 metres in 9.39 seconds at the World Humanoid Robot Games in Beijing, technically beating Usain Bolt’s 9.58-second world record.

The catch: the robots were not exactly graceful about it. Tiangong slammed into a stopping mat and stumbled into the sidelines, while rival robot Lightning finished in 9.47 seconds before collapsing and getting carried away. So yes, the robots are faster now. They just haven’t quite nailed the whole stopping part.

LinkedIn users have had ENOUGH

LinkedIn says more than one million people have already used its new “Seems like AI slop” button since it launched on 30 July.

The button sits in the three-dot menu on posts and lets users flag content that looks heavily AI-generated.

It launched a few weeks after AI detector Pangram found that 41% of LinkedIn’s long-form posts were flagged as fully AI-generated.

LinkedIn has also improved its systems for spotting AI-heavy posts and removed its AI-powered “enhance your post” feature.

Chief product officer Hari Srinivasan says users are now seeing 40% less content LinkedIn classifies as AI slop than they were a few weeks ago.

Here’s what you should know:

1M+ people have used the AI slop button.
41% of long-form posts were flagged as fully AI-generated in Pangram’s analysis.
LinkedIn says views of AI slop are down 40%.

“Thrilled to announce” that this wasn’t written by you

The platform is also adding feedback for people whose posts get flagged.

They may now see a message saying that some members thought their post seemed AI-generated.

LinkedIn says the goal is to give people useful feedback rather than assume they meant to post low-quality AI content.

Earlier this year, it also said it would cut down on comments posted at scale with little or no human input.

Imagine spending 45 minutes prompting an AI to make you sound “authentic and insightful” only for Dave from procurement to hit the slop button before lunch. Devastating. - MG

Mindstream Picks

Health: Europe’s ongoing salmonella outbreak may be linked not only to imported eggs but also chicken products, with more than 200 UK cases confirmed so far.

Gaming: An upcoming PC game turns “finding a needle in a haystack” into a literal challenge, giving players just a one-in-five-million chance of winning on their first try.

Movies: Kit Harington will play Gilderoy Lockhart in Season 2 of HBO’s Harry Potter series, replacing Nicholas Hoult after previously voicing the character in Audible’s audiobooks.

Don’t Miss: Tesla has reportedly discontinued its Solar Roof, ending the solar-powered roofing tile project nearly a decade after it was first unveiled. The product struggled to gain traction, with high costs, installation challenges, and deployments falling far short of Tesla’s original ambitions.