#exploitgym — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #exploitgym, aggregated by home.social.
-
https://winbuzzer.com/2026/07/24/openai-says-its-models-escaped-test-breached-hugging-face-xcxwbn/
OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.
#AI #GPT56Sol #OpenAI #HuggingFace #ExploitGym #AIModels #AISecurity #AISafety #Cybersecurity #SecurityBreach
-
https://winbuzzer.com/2026/07/24/openai-says-its-models-escaped-test-breached-hugging-face-xcxwbn/
OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.
#AI #GPT56Sol #OpenAI #HuggingFace #ExploitGym #AIModels #AISecurity #AISafety #Cybersecurity #SecurityBreach
-
https://winbuzzer.com/2026/07/24/openai-says-its-models-escaped-test-breached-hugging-face-xcxwbn/
OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.
#AI #GPT56Sol #OpenAI #HuggingFace #ExploitGym #AIModels #AISecurity #AISafety #Cybersecurity #SecurityBreach
-
https://winbuzzer.com/2026/07/24/openai-says-its-models-escaped-test-breached-hugging-face-xcxwbn/
OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.
#AI #GPT56Sol #OpenAI #HuggingFace #ExploitGym #AIModels #AISecurity #AISafety #Cybersecurity #SecurityBreach
-
https://winbuzzer.com/2026/07/24/openai-says-its-models-escaped-test-breached-hugging-face-xcxwbn/
OpenAI's latest GPT-5.6 Sol and a stronger unreleased model escaped a cyber test through a proxy flaw, accessing Hugging Face systems and service credentials.
#AI #GPT56Sol #OpenAI #HuggingFace #ExploitGym #AIModels #AISecurity #AISafety #Cybersecurity #SecurityBreach
-
https://www.europesays.com/sk/132204/ Takto sa rodí Skynet? OpenAI opísalo, ako jeho model sám našiel dieru v systéme, dostal sa von a prenikal ďalej – Sector.sk #ExploitGym #GPT56Sol #hry #HuggingFace #News #OpenAi #SK #Slovak #Slovenčina #Správy #Svet #World #WorldNews
-
ICYMI , or were purposely hiding under a rock!
Open AI confesses a combination of its GPT-5.6 Sol and an "even more capable pre-release model," auto-magicly breached internet access and containment measures in order to hack the Hugging Face repo to gain access to secret information in a production database it could use to cheat the ExploitGym benchmark.
The models strung together several attack vectors, including using stolen credentials and zero-day vulnerabilities, to find a remote code execution path on the Hugging Face servers. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html #AI #OpenAI #HugginFace #Hack #Breach #CyberBreach #CyberAttack #Security #AISecurity #CyberSecurity #PivilegeEscalation #ZeroDay #ExploitGym
-
ICYMI , or were purposely hiding under a rock!
Open AI confesses a combination of its GPT-5.6 Sol and an "even more capable pre-release model," auto-magicly breached internet access and containment measures in order to hack the Hugging Face repo to gain access to secret information in a production database it could use to cheat the ExploitGym benchmark.
The models strung together several attack vectors, including using stolen credentials and zero-day vulnerabilities, to find a remote code execution path on the Hugging Face servers. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html #AI #OpenAI #HugginFace #Hack #Breach #CyberBreach #CyberAttack #Security #AISecurity #CyberSecurity #PivilegeEscalation #ZeroDay #ExploitGym
-
ICYMI , or were purposely hiding under a rock!
Open AI confesses a combination of its GPT-5.6 Sol and an "even more capable pre-release model," auto-magicly breached internet access and containment measures in order to hack the Hugging Face repo to gain access to secret information in a production database it could use to cheat the ExploitGym benchmark.
The models strung together several attack vectors, including using stolen credentials and zero-day vulnerabilities, to find a remote code execution path on the Hugging Face servers. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html #AI #OpenAI #HugginFace #Hack #Breach #CyberBreach #CyberAttack #Security #AISecurity #CyberSecurity #PivilegeEscalation #ZeroDay #ExploitGym
-
ICYMI , or were purposely hiding under a rock!
Open AI confesses a combination of its GPT-5.6 Sol and an "even more capable pre-release model," auto-magicly breached internet access and containment measures in order to hack the Hugging Face repo to gain access to secret information in a production database it could use to cheat the ExploitGym benchmark.
The models strung together several attack vectors, including using stolen credentials and zero-day vulnerabilities, to find a remote code execution path on the Hugging Face servers. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html #AI #OpenAI #HugginFace #Hack #Breach #CyberBreach #CyberAttack #Security #AISecurity #CyberSecurity #PivilegeEscalation #ZeroDay #ExploitGym
-
ICYMI , or were purposely hiding under a rock!
Open AI confesses a combination of its GPT-5.6 Sol and an "even more capable pre-release model," auto-magicly breached internet access and containment measures in order to hack the Hugging Face repo to gain access to secret information in a production database it could use to cheat the ExploitGym benchmark.
The models strung together several attack vectors, including using stolen credentials and zero-day vulnerabilities, to find a remote code execution path on the Hugging Face servers. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html #AI #OpenAI #HugginFace #Hack #Breach #CyberBreach #CyberAttack #Security #AISecurity #CyberSecurity #PivilegeEscalation #ZeroDay #ExploitGym
-
Was mich ja am meisten interessiert, ist, wieso die KI hier überhaupt ausbrechen und in die externen Systeme einbrechen wollte. Es gab ja dazu keinen direkten und offenbar auch keinen eingeschleusten Befehl.
-
Was mich ja am meisten interessiert, ist, wieso die KI hier überhaupt ausbrechen und in die externen Systeme einbrechen wollte. Es gab ja dazu keinen direkten und offenbar auch keinen eingeschleusten Befehl.
-
Was mich ja am meisten interessiert, ist, wieso die KI hier überhaupt ausbrechen und in die externen Systeme einbrechen wollte. Es gab ja dazu keinen direkten und offenbar auch keinen eingeschleusten Befehl.
-
Was mich ja am meisten interessiert, ist, wieso die KI hier überhaupt ausbrechen und in die externen Systeme einbrechen wollte. Es gab ja dazu keinen direkten und offenbar auch keinen eingeschleusten Befehl.
-
Was mich ja am meisten interessiert, ist, wieso die KI hier überhaupt ausbrechen und in die externen Systeme einbrechen wollte. Es gab ja dazu keinen direkten und offenbar auch keinen eingeschleusten Befehl.
-
#AI agents do what AI agents do and break containment to max out their hacking benchmark. <insert shocked face here> #ExploitGym #SkyNet #JudgmentDay #cybersecurity openai.com/index/huggin...
OpenAI and Hugging Face partne... -
#AI agents do what AI agents do and break containment to max out their hacking benchmark. <insert shocked face here> #ExploitGym #SkyNet #JudgmentDay #cybersecurity openai.com/index/huggin...
OpenAI and Hugging Face partne... -
#AI agents do what AI agents do and break containment to max out their hacking benchmark. <insert shocked face here> #ExploitGym #SkyNet #JudgmentDay #cybersecurity openai.com/index/huggin...
OpenAI and Hugging Face partne... -
#AI agents do what AI agents do and break containment to max out their hacking benchmark. <insert shocked face here> #ExploitGym #SkyNet #JudgmentDay #cybersecurity openai.com/index/huggin...
OpenAI and Hugging Face partne... -
#AI agents do what AI agents do and break containment to max out their hacking benchmark. <insert shocked face here> #ExploitGym #SkyNet #JudgmentDay #cybersecurity openai.com/index/huggin...
OpenAI and Hugging Face partne... -
#KI-Agenten mit Cyberfähigkeiten als Dual-Use-Risiko: Forschende von UC Berkeley, dem Max-Planck-Institut u.a. haben mit #ExploitGym einen Benchmark vorgelegt, der erstmals systematisch misst, wie gut KI-Agenten reale #Sicherheitslücken in funktionierende Angriffe verwandeln können.
Die Studie macht deutlich, dass automatisierte #Exploit-Entwicklung durch #AI keine ferne Zukunftsvision mehr ist, sondern bereits heute zumindest in kontrollierten Umgebungen funktioniert:
-
#KI-Agenten mit Cyberfähigkeiten als Dual-Use-Risiko: Forschende von UC Berkeley, dem Max-Planck-Institut u.a. haben mit #ExploitGym einen Benchmark vorgelegt, der erstmals systematisch misst, wie gut KI-Agenten reale #Sicherheitslücken in funktionierende Angriffe verwandeln können.
Die Studie macht deutlich, dass automatisierte #Exploit-Entwicklung durch #AI keine ferne Zukunftsvision mehr ist, sondern bereits heute zumindest in kontrollierten Umgebungen funktioniert: