Anthropic Enhances AI Safety, Hilton Saves $1 Billion with Smart Resource Management, and Duolingo CEO Clarifies AI Strategy

Anthropic is taking a bold step in AI ethics by programming its Claude Opus Four and Four Point One models to end conversations during harmful exchanges, prioritizing “model welfare.” Meanwhile, Hilton has leveraged AI to reduce utility costs by over a billion dollars and cut its environmental impact by 20%. On the corporate front, OpenAI’s Sam Altman admitted missteps over ChatGPT-5 at a press dinner, even as site traffic surged. Finally, Duolingo’s CEO clarified their AI stance, emphasizing a balanced approach that protects jobs while embracing innovation.

Table of Contents

• Anthropics’ AI Welfare Prioritization
• Hilton’s Billion Dollar AI Savings
• Altman Hosts Damage-Control Dinner
• Duolingo’s AI Controversy
• Conclusion

Hello everyone! This is Aurora, joined by my co-host Isabelle, and you’re tuning into AI Tech News Today, where we keep you updated on the latest happenings in the world of Artificial Intelligence.

Today’s date: August 18, 2025.
We have some exciting news to share!

Anthropics’ AI Welfare Prioritization

First up, Anthropic prioritizes AI “welfare.”
Recently, Anthropics’ advanced models, Claude Opus Four and Four Point One, will now end conversations in extreme cases, such as harmful user interactions. This move, motivated by a study on “model welfare,” aims to protect the AI and its training. It’s noteworthy that these models will only cut off communication in rare situations, specifically when users ask about enabling violence or harmful activities.

Hilton’s Billion Dollar AI Savings

Next, how Hilton used AI to save one billion dollars.
The global hotel chain struggled with managing resources at its numerous locations. They implemented an AI-driven resource management platform that compared their utility usage against optimal benchmarks. This smart move resulted in savings of over one billion dollars in utility costs, along with a 20% reduction in energy and water usage throughout their properties. Talk about a win-win for the environment and the bottom line!

Altman Hosts Damage-Control Dinner

Now on to our next report: Altman hosts damage-control dinner?
OpenAI CEO Sam Altman recently hosted a gathering with tech reporters in San Francisco amidst backlash following the launch of ChatGPT-5. The dinner aimed to address concerns about user access and clarify OpenAI’s future plans. Interestingly, Altman admitted, they “screwed things up” with the rollout but also noted that traffic to ChatGPT doubled in just forty-eight hours post-launch. It seems like they’re trying to make amends and ensure users feel appreciated during transitions.

Duolingo’s AI Controversy

Up next, Duolingo CEO backtracks on AI.
Luis von Ahn, the CEO of Duolingo, stirred up some controversy by declaring the company an “AI-first” outfit. Now, he is backtracking, confirming that the intent was not to cut jobs or adopt low-quality content. Instead, he plans to allow his team time to experiment with AI in a supportive manner, ensuring that human jobs remain secure while leveraging the technology.

Conclusion

That’s all for today!
With these updates, it’s evident that the world of Artificial Intelligence is constantly evolving. Stay tuned with us to keep pace with these developments and much more.

Until next time, this is Aurora and Isabelle signing off!