OpenAI's IPO: Altman Hints at "Within a Year," but the Timeline Is Slipping
OpenAI CEO Sam Altman has told employees that the company expects to carry out an initial public offering (IPO) "within a year." However, reports indicate that the plan has been pushed back from the originally anticipated timeline.
Altman has long taken a cautious stance on going public, exploring ways to balance an IPO with preserving the company's nonprofit governance structure. His latest remarks were internal, and no specific date has been announced yet. OpenAI is already valued at hundreds of billions of dollars, and if the IPO materializes, it could become one of the largest technology debuts in recent years.
Visa Integrates Payments into ChatGPT, Letting AI Agents Shop on Your Behalf
Visa has partnered with OpenAI to announce a system that enables AI agents to purchase and pay for products through ChatGPT. By connecting Visa's payment network directly to ChatGPT, users can complete a purchase simply by telling the AI to "buy X."
The move signals that AI agents are now seriously entering the territory of real economic activity, beyond simply providing information. Mastercard is also reportedly pursuing similar initiatives, and the convergence of payment infrastructure and AI is accelerating. At the same time, there is ongoing debate over the need for safeguards against AI-driven fraud and unintended purchases.
Anthropic Research: AI Can Build Exploits from Security Patches Within Hours
Research released by Anthropic has revealed that AI models can analyze publicly released security patches and construct a working exploit in just a few hours. Traditionally, exploiting a vulnerability typically required experts to spend weeks on analysis.
The findings underscore the dual-edged nature of AI in cybersecurity. While defenders are also making progress in leveraging AI for early vulnerability discovery, critics warn that attackers' use of AI could advance at a pace that outstrips defenses.
Federal Court: "AI Search Is Not Essential," Ruling Against Google
In litigation related to Google's monopoly in the search market, a federal court has determined that "AI-powered search functionality is not indispensable." The ruling is widely interpreted as a check on Google citing its provision of AI search features as a justification for its monopoly.
This is the first major judicial decision to weigh the social value of AI search against its impact on market competition, and it could influence the future shape of the search market.
Trump: "AI Companies Should Give Back to Society," Expects Voluntary Agreements
President Donald Trump has indicated that he wants AI companies to "give back" a portion of their profits to society. The president has expressed optimism that AI companies will voluntarily agree to such an arrangement.
The specifics remain unclear, but the move is seen as an effort to demand some form of contribution from the industry in response to AI-driven job displacement and copyright issues. Whereas AI regulation debates have so far focused mainly on safety, this new approach — economic give-back — is drawing attention.
Fine-Tuning GPU Clock Frequencies Could Drastically Cut Energy Consumption in LLM Training
According to IEEE Spectrum, new research shows that slightly lowering GPU clock frequencies can dramatically reduce the energy consumed during LLM training. The impact on performance is said to be minimal, while power efficiency improves substantially.
Large-scale LLM training requires enormous amounts of electricity, making environmental impact and cost major challenges. If this technique gains traction, it could help curb data center power consumption while maintaining training efficiency.