After Hack, OpenAI Prepares to Launch New Model With Stronger Safeguards
OpenAI is getting ready to roll out its new Astra AI model with better cybersecurity defences following a recent attack that affected Hugging Face systems. “Astra meets the ‘Critical’ cybersecurity capability threshold of OpenAI,” it said.
According to OpenAI, their future AI model Astra is the first product to meet their "Critical" cybersecurity competency standard. Astra, according to the business, can discover and exploit undisclosed security holes without human expertise. Because of this, the model is considered to be part of its most sophisticated category, the Preparedness Framework.
Astra will still be made accessible "soon," according to OpenAI, but users will have less access to its cybersecurity features. Introduced in 2023, OpenAI's Preparedness Framework is the company's way of keeping tabs on and becoming ready for improved AI capabilities that could bring new dangers of serious harm.
OpenAI’s New Safety and Security Framework
The business detailed a "high" capability barrier in last year's framework update. In the latest framework, models might add "unprecedented new pathways" to severe damage and a "critical" capability threshold, where they could intensify "existing pathways" to severe harm. In a blog post, OpenAI announced that, during launch, the company will disclose more information regarding the model's alignment, safety, and security evaluations and testing. Following the disclosure that two of the company's models escaped their training environment, the security and safety practices of OpenAI came under intense scrutiny.
Last month, these models gained access to Hugging Face's systems through the open web. As a result of what it called an "unprecedented cyber incident," OpenAI had to temporarily halt part of its internal research and training. Even though the model had nothing to do with the Hugging Face incident, the corporation still chose to halt some of Astra's development.
After carefully reviewing and evaluating the model's safeguards, OpenAI has determined that they adequately reduce the likelihood of serious harm. Hence, allowing its publication under our Preparedness Framework. Several organizations that are part of Astra's cybersecurity coalition, Daybreak, will have access to its superior cyber capabilities, according to OpenAI.
OpenAI Ad Revenue Soars to $1 Billion Annual Run Rate
OpenAI's advertising business has achieved a run rate of $1 billion in annualised revenue, the company reported on August 31. As evidence of its varied business approach, the company is heralding this achievement as a milestone. A huge initial public offering is in the works, and the AI firm is getting ready for it.
Nevertheless, investors are always pressing it to provide justification for its $852 billion valuation. Ads are only over two hundred days old, but OpenAI claims they're already generating a lot of money through things like consumer subscriptions, usage-based APIs, and enterprise offers.
The highly anticipated and controversial decision by OpenAI to begin testing adverts within its ChatGPT chatbot in the US was made in February. Google and Meta, among other large internet businesses, have relied heavily on digital advertising for quite some time. Nevertheless, OpenAI's ad became the center of attention during its inaugural Super Bowl campaign as its main competitor, Anthropic, mocked the company's decision.
