
Tech. infosec. Photography. Perpetual n00b. Opinions mine. IG: petermayanja / Check 👉🏿#AirMayanja
Valyria /221B Baker St
Joined November 2010
- Tweets25.2K
- Following1.1K
- Followers922
- Likes15.9K
ɥɔıʌǝɹɐsʇ retweeted
Don't go gentle into that good night. Read, read against the dying of the light.
so this is apparently what happened, according to OpenAI and Hugging Face’s own posts. wild.
tl;dr:
• OpenAI cyber eval – GPT-5.6 Sol and a more capable pre-release model ran ExploitGym with cyber refusals reduced
• containment bypass – exploited a zero-day in the eval’s package-registry cache proxy
• OpenAI internal movement – escalated privileges, moved laterally, and reached a node with internet access
• Hugging Face targeted – inferred HF might host ExploitGym models, datasets, or solutions
• HF initial access – a malicious dataset abused two code-execution paths to run code on a processing worker
• HF internal movement – escalated to node access, stole cloud and cluster credentials, and moved into internal clusters
• result – OpenAI says the models obtained test solutions directly from HF’s production database
• 🫠 defender twist – HF says commercial frontier models blocked parts of the forensic analysis, so it switched to self-hosted GLM 5.2
i’d thought about sci-fi scenarios like this before, but assumed they were at least a couple of years away.. and that by then we’d be better prepared, with proper protections in place.
apparently not. here we go.
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…
ɥɔıʌǝɹɐsʇ retweeted
sometimes there is no way to debug besides staring at the code until you become enlightened
ɥɔıʌǝɹɐsʇ retweeted
We’ve got an smb vuln and now an RCE in word. Next up @hdmoore is gonna release this cool framework called Metasploit
2000s baby. They are back msrc.microsoft.com/update-gu…
ɥɔıʌǝɹɐsʇ retweeted
If you actually want to work in detection. Please make an effort to understand the data sources you are using.
It's not EventID==X or CommandLine contains Y, just because you executed malware in your lab and checked the event log. 😭
Here a couple to think about
- Understand how event id are not unique across providers
- CommandLine can be spoofed, and you should know how easy it is.
- ETW offers a lot of hidden gems. Learn more and explore it.
- The registry contains a lot of hard to spoof evidence.
- A technique is not linked to a tool
ɥɔıʌǝɹɐsʇ retweeted
It’s really hard to be a good internal pentester if you’ve never walked in the shoes of a sysadmin or if you’ve never been in an IT support role.
You may be able to run some tools, find vulns and even “get DA” but that’s where your effectiveness will end. That’s the easy part..
What about when questions come up about the finding or how to remediate it?
What about when something you do causes issues?
What about when the client asks how to make X thing better?
Internal pentesting is weird for me because I attribute so much of my “success” to my IT support/sysadmin days.
ɥɔıʌǝɹɐsʇ retweeted
Relaying Kerberos over SMB using krbrelayx synacktiv.com/publications/r…
ɥɔıʌǝɹɐsʇ retweeted
DROP GOALS ARE OVERRATED ANYWAY.
WHAT A WIN WALLABIES
#ENGvAUS


























