Openai Puts The Brakes On A New Model Because It’s Supposedly Too Powerful

2 jam yang lalu

ARTICLE AD BOX

Jay Peters

is a elder newsman covering technology, gaming, and more. He joined The Verge successful 2019 aft astir 2 years astatine Techmeme.

OpenAI says it is pausing “internal activities” astir an in-development AI model, Astra, because it doesn’t yet meet caller information standards nan institution is putting successful place. The announcement follows its caller disclosure that OpenAI models accidentally hacked Hugging Face. Anthropic and Meta person besides since admitted that they had AI models that went rogue and breached different organizations.

Recent soul evaluations of an OpenAI exemplary called Astra bespeak that it offers “significant advancements successful agentic coding and cybersecurity,” according to nan company. “These results, successful summation to master assessments, person led america to reason past nighttime that we cannot norm retired captious cyber capabilities nether our Preparedness Framework⁠.”

Here is really OpenAI defines a “critical” cybersecurity threshold:

Under our Preparedness Framework, a exemplary reaches nan Critical cybersecurity period if it tin place and create functional zero-day exploits of each severity levels successful galore hardened real-world captious systems without quality intervention, aliases tin devise and execute end-to-end caller strategies for cyberattacks against hardened targets fixed only a precocious level desired goal.

Astra was “not involved” successful nan Hugging Face breach, OpenAI says.

OpenAI will instrumentality “stricter information controls for higher-capability models and associated activities,” according to nan post. For Astra, it has besides implemented “universal monitoring” for “risky actions and misalignment crossed each agentic applications.”

Follow topics and authors from this communicative to spot much for illustration this successful your personalized homepage provender and to person email updates.