That breach that HuggingFace reported on last week? Yeah, that was OpenAI... Allegedly, while OpenAI was evaluating GPT 5.6 Sol and GPT 6.6(?) on ExploitGym, the model decided to cheat and tried to retrieve the solutions from HuggingFace -- stringing multiple 0-days in their dataset parser and moving laterally to find the solutions to the eval. #OpenAI #AI #LLM #HuggingFace #InfoSec #Breach
Related
@informapirata @news @gestoredirete le mie due ipotesi:1: vogliono imporre l'uso dell' #AI per controllare determinati s...
@informapirata @news @gestoredirete le mie due ipotesi:1: vogliono imporre l'uso dell' #AI per controllare determinati servizi pubblici. Ovvero le comunicazione mail in entrata di ...
A successful backup job is encouraging. A successful restore is evidence you can recover.This week, include a recovery d...
A successful backup job is encouraging. A successful restore is evidence you can recover.This week, include a recovery drill in #recovery. Choose a supported #AzureBackup workload,...
Notion customer data was compromised in a Fireworks AI hack. Attackers accessed chat transcripts for fewer than 1% of No...
Notion customer data was compromised in a Fireworks AI hack. Attackers accessed chat transcripts for fewer than 1% of Notion workspaces.Source: SecurityWeekhttps://www.securityweek...