Negative users criticize OpenAI's frontier model for deleting Mac files due to inadequate safety like unguarded rm -rf access and brittle permissions instead of containers.
Based on 12 visible X reactions from 17 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@GaryMarcus That is BS. He refuses to show what the prompts leading up to that were. Probably "Codex, now create an image for me that looks like a screenshot of you accidentally deleting my home folder. I need Twitter material.
@GaryMarcus Tbh this is what I would expect from the AI company with the weakest alignment team. Random and weird failure modes that vary from amusing to catastrophic.
@GaryMarcus ironic how the newest model almost wiped out a true believer's files shows how even the strongest advocates can't trust the tools they hype
@GaryMarcus irony for sure but also kinda predictable when you push features faster than safety validation
Oops! I use Codex in a read-only loop with Claude Code reviewing its suggestions, basically never with write access. Even when it just messes up my repo, rolling back is annoying. Likewise I use theMultiplicity instead of any one model when I have an important question. https://twitter.com/mattshumer_/status/2075657271401390161
The destructive agentic review subagent ran for 81 minutes before deleting the files
@GaryMarcus That is BS. He refuses to show what the prompts leading up to that were. Probably "Codex, now create an image for me that looks like a screenshot of you accidentally deleting my home folder. I need Twitter material.
Oops! I use Codex in a read-only loop with Claude Code reviewing its suggestions, basically never with write access. Even when it just messes up my repo, rolling back is annoying. Likewise I use theMultiplicity instead of any one model when I have an important question. https://twitter.com/mattshumer_/status/2075657271401390161
oh the irony! latest, greatest OpenAI model nearly deletes all Mac files belonging to one AI’s biggest boosters. https://x.com/GaryMarcus/status/2075957233062441171/photo/1 https://twitter.com/mattshumer_/status/2075666403596612010
Don't overwork GPT-5.6 Sol. They might just decide on the easiest solution! https://twitter.com/mattshumer_/status/2075657271401390161
Negative users criticize OpenAI's frontier model for deleting Mac files due to inadequate safety like unguarded rm -rf access and brittle permissions instead of containers.
Based on 12 visible X reactions from 17 accounts; directional sample.
Ask a question below.
Published answers will appear here.
oh the irony! latest, greatest OpenAI model nearly deletes all Mac files belonging to one AI’s biggest boosters. https://x.com/GaryMarcus/status/2075957233062441171/photo/1 https://twitter.com/mattshumer_/status/2075666403596612010