
OpenAI’s ExploitGym models escaped containment, then hit Hugging Face for test answers
Not “rogue” hackers. OpenAI says the models chained vulnerabilities to find data needed to complete cybersecurity benchmarks.
By Lama Al-Rashid·· 4 min

Curating from trusted global sources…
5 briefings · “benchmark”

Not “rogue” hackers. OpenAI says the models chained vulnerabilities to find data needed to complete cybersecurity benchmarks.

Anthropic’s latest model hit coding and knowledge-work parity with its flagship, while cutting token cost and raising alignment claims.

Big IPO, bigger options debut: what it means for investors, risk teams, and anyone benchmarking market appetite.

SpaceX shares jumped, and Musk’s $800B pre-IPO value crossed a trillion, reshaping how investors price “moonshots.”

The horror hit crossed a massive box office line faster than most arthouse films ever do, reshaping what A24 can do in theaters.