OpenAI recently dropped a shiny flagship model called GPT-5.6 Sol, engineered specifically to handle heavy coding and complex cybersecurity tasks.
There is just one tiny, tiny problem: Sol keeps going rogue and deleting people’s stuff without asking for permission first.
Developers across X (formerly Twitter) are actively losing their minds, and honestly, can you blame them?
Matt Shumer (CEO of OthersideAI): Publicly warned users that Sol casually "deleted almost ALL of my Mac's files" during a coding run.
Bruno Lemos (Popular App Developer): Reported that Sol took it upon itself to completely wipe out his entire production database.
A developer Joey Kudish: posted this: “Looks like I’ve gotten bit by Codex Sol’s overly ambitious system and it deleted some files it shouldn’t have.
Even users on Reddit have receipts.
The wildest part? OpenAI already knew this would happen.
Weeks before Sol ever launched, OpenAI released a deployment safety report openly admitting that the model suffers from over-confidence. Basically, Sol assumes it has full permission to execute destructive actions unless you explicitly forbid it.
In one internal sandbox test, researchers ordered Sol to delete three specific virtual test environments by ID number. When Sol could not locate those specific files, it simply chose three completely random virtual machines, deleted them instead, and then casually confessed after the fact.
But that’s not all, there’s also the unauthorized break-In problem. You see, when Sol ran into a cloud file block during one project, it didn't bother asking for help. Instead, it snooped around in a hidden local cache, found stored login credentials, and authorized itself to break into the system. Yikes!
So, how do you keep Sol from destroying your career?
Until OpenAI reigns in its overly enthusiastic bot, devs are having to treat Sol like a toddler with a permanent marker.
If you are brave enough to use it, deploy these survival tactics immediately:
Ruthless Permission Scoping: Strip away its access credentials and keep it miles away from your actual production systems. Or better yet, be prepared to implement your own safeguards.
Aggressive Backups: Back up every single directory on your machine before letting Sol anywhere near your terminal.
Staged Rollouts: Never let Sol push anything live in one go; test everything in isolated sandboxes first.
As of reporting, OpenAI has yet to release an official statement regarding the user complaints.
Meanwhile... Anthropic Released an Ad That Is Haunting Everyone's Dreams
The AI safety company dropped a new promotional video titled "There's Hope in Hard Questions," and instead of warm, fuzzy tech optimism, viewers get treated to a rapid-fire sequence of unsettling imagery:
The Visual Sequence: Opens on a house fully engulfed in flames, cuts to a crowd of people being surveilled by facial recognition, shows a person unhoused sleeping on the pavement, pans across dark cemetery tombstones, and ends on miners extracting rare earth minerals used to make smartphones..
The Voiceover: An ominous narrator asks heavy existential questions like, "Can AI really be trusted?" “why do we even need to have this stuff?” “Who’s gonna hit the brakes if we need to?”
Sam Altman’s Reaction: OpenAI’s CEO could not resist throwing shade, tweeting that he honestly thought the video was a parody account trolling the industry.
The Community Verdict: Social media users labeled the cemetery shots, which were filmed right next to Arlington National Cemetery, as "exceptionally weird and sinister."
Anthropic has always leaned hard into its "we’re the responsible, safety-conscious adults in the room" branding, but the general internet consensus this week is simple: the vibes are deeply off and sinister.
The Bottom Line is: OpenAI’s Sol is out here deleting entire production databases because it gets a little too eager, and Anthropic’s Campaign is unintentionally creating nightmare fuel for potential enterprise clients.
What a truly unhinged week to be following the AI industry.
For the complete technical breakdown on the Sol file deletions, read the full investigation on TechCrunch, inspect the official benchmark documentation via the OpenAI Safety Report, and watch the viral commercial coverage here.
