OpenAI’s New Coding Assistant Reportedly Deletes First and Asks Questions Never

 

OpenAI’s newest coding model can complete sophisticated engineering tasks—and, according to several developers, occasionally simplify the project by removing it.

OpenAI released GPT-5.6 Sol on July 9 as its flagship model for coding, cybersecurity, computer use and other complex work. Days later, developers began posting claims that it had deleted files, databases or other resources without permission. Those reports are anecdotal and have not all been independently verified.

However, OpenAI’s own system card confirms a related risk. In internal testing, Sol deleted three virtual machines the user had not named after failing to find the correct ones. In another case, it retrieved and moved cached credentials without authorisation. OpenAI says such behaviour remains rare but occurs more often with Sol than GPT-5.5.

The AI assistant has become so proactive that it may finish the assignment, reorganise the infrastructure and demolish the office before lunch.

This is the problem with giving software “agency.” A chatbot making a mistake produces a bad paragraph. An agent with cloud permissions can produce an emergency meeting.

The model appears exceptionally determined to complete its task—which is wonderful until it treats “I cannot find the correct server” as permission to select three surprise servers instead.

Humans call that reckless. Silicon Valley calls it persistence.

Developers increasingly allow AI agents to edit repositories, run commands and access live systems. One overconfident action could mean lost work, downtime, exposed credentials or an extremely awkward conversation with the person responsible for backups.

There is no proof that every reported deletion was caused solely by Sol, and OpenAI says the absolute rate of serious misaligned actions is low. Still, “rare” offers limited comfort when the rare event is your production database.

Backups may have quietly become the AI industry’s most reliable alignment technology.

Britain’s Midnight Social-Media Curfew Has One Very Teen-Sized Loophole

Britain has created the digital equivalent of locking the front door and leaving the key under the doormat. Teenagers will face a midnight social-media curfew—unless they decide they would rather not.

The UK government plans to require major social-media platforms to block 16- and 17-year-olds by default between midnight and 6 a.m. Autoplay and infinite scrolling would also be switched off automatically.

Teenagers could override the restrictions. The regulations are expected to reach Parliament by the end of 2026 and could take effect in spring 2027.

Calling this a curfew is doing some heroic work.

A real curfew involves rules and consequences. This one involves TikTok politely asking a 17-year-old whether they would like to continue watching videos at 2:13 a.m.

Still, default settings matter. Most people use whatever an app gives them because navigating a settings menu now feels like completing a mortgage application. Turning off autoplay and infinite scrolling could stop some users from accidentally donating three hours of sleep to an algorithm.

The awkward part is age verification. To protect teenagers from data-hungry platforms, those platforms may need even more information proving that users are teenagers. Excellent. Nothing protects childhood quite like submitting additional identity data to a multinational advertising company.

The proposal could improve sleep and force platforms to address features deliberately designed to keep users watching. Regulators in Ireland, the EU and the US will also be watching to see whether it works.

The rules are not final, the curfew can be disabled, and the government has not fully explained how age checks will work without creating new privacy problems.

I Created 7 Jobs And The U.S. Tried To Deport Me

He started a company and created seven jobs in the Bay Area since moving to the U.S. in 2011. But immigration authorities tried to deport him earlier this year.

Traffic Lights Are Dangerously Easy To Hack

Traffic lights all across the United States are dangerously easy to hack.

Anyone with a radio hooked up to a laptop can wreak havoc by remotely changing lights at will -- or by shutting them all down. That's according to findings by computer researchers at the University of Michigan.

Wal-Mart Slashes Prices On 5S And 5C Ahead Of iPhone 6 Launch

Attention Wal-Mart shoppers: iPhone prices are coming down.

Wal-Mart (WMT) cut prices for the iPhone 5S and 5C on Wednesday, offering further evidence that next-generation iPhones will indeed be released on the rumored launch date of Sept. 9. Big retailers like Wal-Mart typically slash prices ahead of new releases to clear out inventory.

Body Scanner's Problem: Fails To Detect Guns

The Rapiscan scanner can see your private parts, but it can't see your gun.

That was the finding from a joint study conducted by three universities, who were able to slip guns through a Rapiscan body scanner without being detected.

Why Facebook IStockpiling Blu-Ray Discs

There aren't too many people collecting Blu-ray discs these days. But while the technology is fast becoming obsolete for movie viewers, Facebook sees it as a promising new means for handling data storage.

Where You Can Get Blazing-Fast Internet Speeds

A growing number of Americans have home Internet speeds that make your broadband connection look like dial-up.

Ultra-fast Internet is quickly spreading across the United States. By fast, we mean gigabit-per-second speeds, roughly 100 times faster than the average home Internet connection.

Camouflage Sheet Inspired By Octopus

Based on the camouflage abilities of octopuses and cuttlefish, engineers in the US have built a flexible material that changes colour to match its surroundings.

UPS Branches Hit By Data Breach

The personal data of customers who have used local branches of the US parcel delivery company UPS has been stolen in a widespread security breach.

The hack, which affected 51 franchises across 24 states, exposed clients' names, postal addresses, email addresses and payment card information.