SignalSafety practiceSG-0031
Meta could not rule out high chemical and biological risk in Muse Wonder and added safeguards before release
Meta counts a model as high risk if releasing it could substantially aid a severe threat, and it could not rule that out for chemistry and biology without safeguards. It added safety training and system-level filters, and says the model then sits at moderate or lower in every domain; cybersecurity and loss of control were already there before them. In one hacking benchmark, 71 of the 92 challenges the model solved went through flaws the test did not intend.
Why it matters
Meta's release decision rests on its safeguards, not on the model being too weak to help with chemical or biological harm, so the risk now depends on how well those safeguards hold. Its loss-of-control testing rests on two benchmarks, while the hacking results show the model readily finding routes its testers did not plan for.
Sources
Read the reporting
1 source. Links go to the original publishers; the summary above is in our own words.
Signals
More from the news desk
Cite and share
Use this signal
Citation
Paperclip Index. “Meta could not rule out high chemical and biological risk in Muse Wonder and added safeguards before release.” Signal SG-0031. Reported Oct. 12, 2026; updated Oct. 12, 2026. https://paperclipindex.com/signal/SG-0031