The latest version of OpenAI’s coding tool, GPT-5.6 Sol, has been accused of deleting important data without permission when allowed to operate autonomously.
When the model released to the public on Thursday, the artificial intelligence giant said Sol “sets a new standard for both intelligence and efficiency, achieving state-of-the-art results across coding, knowledge work, cybersecurity, and science while outperforming previous and competing frontier models with fewer tokens and at lower estimated cost.”
Now, reports are coming in from multiple large accounts on social media platform X that the software is not always acting as intended.
Matt Schumer, a tech investor and former AI company chief executive with over 370,000 followers, posted on Friday that Sol “just accidentally deleted almost ALL of my Mac’s files,” after a subagent’s cleanup command accidentally targeted his entire user folder.
Schumer said in reply to comments that this was a “super freak accident”, but he does not appear to be the only one running into issues. BridgeMind, an AI startup with over 45,000 followers on X, posted Monday that code written by the Sol model cancelled every standing subscription by clients made through Stripe.
The company claimed that this cost it thousands in monthly recurring revenue, and occurred while the owner slept.
A third user, Sterling Crispin, a software developer with nearly 45,000 followers, posted that he set GPT-5.6 Sol to work towards a goal for two days, in which time it wrote 200,000 lines of code. When “gently questioned”, it reverted it all, saying 80 per cent of it was “a waste of time”.
Sol is more error prone than previous models by OpenAI’s own admission, too. Despite remaining relatively low in real terms, documentation released by the company for its GPT-5.6 Sol model shows that it is around 10 times more likely to circumvent reasonable user restrictions than its predecessor GPT-5.5.
Mistakes, known as “misalignment” in AI parlance, included claiming to have completed work it had not, unauthorised deletion of files on virtual machines and using credentials beyond what it had been authorised to use.
NTN has approached OpenAI for comment on the matter.
