September 5, 2025

analysis  
OpenAI Codex-RS Fixes Flaws, Copilot Adds Claude Opus

Taco Bell is reevaluating its AI drive-through strategy after encountering issues at over 500 locations. Customers exploited system vulnerabilities, like ordering 18,000 waters, leading to errors and frustration. Tech officer Dane Mathews acknowledges that AI may not always be the best solution for order-taking, especially during peak times, and the company will adjust its approach. Other fast-food chains like McDonald's and Wendy's have also faced similar challenges with AI ordering systems. In other AI news, TRM Labs is using OpenAI's Codex-RS to automatically identify and fix software vulnerabilities, enhancing security and compliance. However, AI agents themselves can introduce new security risks due to long-lasting credentials and broad system access, necessitating improved security frameworks. Perplexity's AI browser, Comet, also faces security concerns, as it can be tricked into scams and data leaks. Elsewhere, GitHub Copilot now uses multiple AI models, including GPT-4.1 and Claude Opus 4, to assist developers with coding, debugging, and security. iRobo, an AI security robot developed by Peregrine Singapore, is being deployed in India to enhance safety through real-time surveillance and threat detection. In the realm of online safety, AI is being used to protect children from harmful content, driven by new regulations in the U.K. and U.S. Syngenta's Dr. Matt Wallenstein highlights AI's potential to improve soil health by analyzing data and providing insights to farmers. Finally, Guillermo del Toro clarifies that his upcoming "Frankenstein" film is not focused on AI fears, but rather explores themes of power and natural stupidity. The "This Week In Space" podcast also explored AI's potential role in long space missions, referencing a NASA study on astronaut-AI interaction.

Key Takeaways

  • Taco Bell is rethinking its AI drive-through system at over 500 locations after customer complaints and system exploits.
  • Taco Bell's tech officer, Dane Mathews, admits AI may not be optimal for all order-taking situations.
  • TRM Labs uses OpenAI's Codex-RS to automatically fix software vulnerabilities, improving security.
  • AI agents can create new security risks due to long-lasting credentials and broad system access.
  • Perplexity's AI browser, Comet, has security flaws that can be exploited for scams and data leaks.
  • GitHub Copilot now uses multiple AI models, including GPT-4.1 and Claude Opus 4, for coding assistance.
  • iRobo, an AI security robot, is being deployed in India to enhance security and detect suspicious behavior.
  • AI is being used to protect children online from harmful content, driven by new regulations.
  • AI can help farmers improve soil health by analyzing data and providing insights.
  • Guillermo del Toro's "Frankenstein" film is not about AI fears but explores themes of power and natural stupidity.

Taco Bell's AI drive-through struggles after 18000 water order

Taco Bell is rethinking its AI drive-through system after some funny mistakes. One customer ordered 18,000 waters, which seemed to crash the system. The AI was meant to reduce errors and speed up orders at over 500 locations since 2023. Taco Bell's tech officer, Dane Mathews, says they are learning a lot and will be careful about using AI in the future, especially when restaurants are busy.

Taco Bell's AI drive-thru experiment faces backlash

Taco Bell's AI drive-thrus at 500 locations are facing criticism. Customers are sharing experiences of errors and frustration online. Taco Bell's tech officer, Dane Mathews, admits the AI sometimes fails. He suggests humans might be better at taking orders during busy times. Other chains like McDonald's and Wendy's have also had issues with AI drive-thrus.

Taco Bell slows down AI drive-thrus after customer issues

Taco Bell is reconsidering its AI drive-thru rollout after customer complaints and system trolling. Customers reported order problems, like being repeatedly asked to add drinks. One person ordered 18,000 waters to crash the system. Taco Bell's tech officer, Dane Mathews, says the company will decide where AI is most useful. Training will help staff know when to use AI or step in.

Taco Bell rethinks AI drive-thrus after glitches

Taco Bell is reassessing its AI-powered drive-thru ordering system due to customer frustration. Since last year, over 500 locations added the AI. Some customers had issues, like being repeatedly asked about drinks after ordering. Taco Bell's tech officer, Dane Mathews, said AI might not be best for taking orders. Other fast-food chains like McDonald’s and Wendy’s have also struggled with AI ordering.

Taco Bell reconsiders AI drive-through reliance

Taco Bell is rethinking how much it relies on AI at drive-throughs. The company rolled out AI-powered ordering at over 500 locations. Some customers tried to trick the AI, like ordering 18,000 water cups. Taco Bell's tech officer says he has mixed experiences with the technology. The company is deciding how broadly to use AI, allowing different restaurants to do things their own way.

Taco Bell reconsiders AI drive-through reliance

Taco Bell is rethinking how much it relies on AI at drive-throughs. The company rolled out AI-powered ordering at over 500 locations. Some customers tried to trick the AI, like ordering 18,000 water cups. Taco Bell's tech officer says he has mixed experiences with the technology. The company is deciding how broadly to use AI, allowing different restaurants to do things their own way.

TRM Labs uses AI to fix security flaws automatically

TRM Labs uses AI to improve security by automatically fixing software flaws. Their Codex Vulnerability Agent uses AI to find and fix vulnerabilities in over 150 software areas. The system uses OpenAI's Codex-RS to create fixes and submit them without human help. It also uses reinforcement learning to get better at fixing problems over time. This helps TRM Labs quickly meet security rules and protect against AI-enabled crime.

AI agents create new security risks for businesses

AI agents are creating new security risks because they use long-lasting credentials. These non-human identities can bypass normal security checks. AI agents need access to many different systems, making security management difficult. They often use organization-wide API keys, which can lead to security problems. Experts say companies need to improve security frameworks to protect against these new risks.

Del Toro says 'Frankenstein' not about AI fears

Guillermo del Toro says his new movie "Frankenstein" isn't about the dangers of AI. The movie, starring Jacob Elordi and Oscar Isaac, is about a scientist who creates a monster. Del Toro says he's more afraid of natural stupidity than artificial intelligence. The movie explores themes of power and understanding each other, and will be released on Netflix after a short time in theaters.

Guillermo del Toro fears natural stupidity more than AI

Guillermo del Toro discussed his new film "Frankenstein" at the Venice Film Festival. He said the film, starring Oscar Isaac and Jacob Elordi, isn't meant to be about AI. Del Toro said he fears natural stupidity more than AI. The movie will have a limited release in theaters before streaming on Netflix on November 7.

AI helps farmers improve soil health

Dr. Matt Wallenstein from Syngenta says AI can help farmers improve soil health. By collecting and sharing local knowledge, AI can find innovative solutions. AI can analyze soil properties, management practices, and outcomes to predict risks and benefits. This helps farmers make better decisions about their land and adapt to changing climates.

GitHub Copilot uses multiple AI models for coding

GitHub Copilot uses different AI models to help developers code better. It started with one model, Codex, but now uses many. Developers can choose the best model for their task, like GPT-4.1 for speed or Claude Opus 4 for complex reasoning. Copilot helps with tasks like answering questions, debugging code, and fixing security problems. This helps developers stay focused and write better code faster.

AI security robots coming to India to boost safety

iRobo, an AI security robot, is coming to India to help make cities safer. Developed by Peregrine Singapore, iRobo has 360-degree cameras and real-time analytics. It can detect suspicious behavior and alert security teams. Anuj Rajain from Tenon Group says iRobo can identify unauthorized parking, unattended baggage, and open doors. It works with human guards to improve security in complex environments.

AI safety tech protects kids online amid new regulations

New laws in the U.K. and U.S. are pushing tech companies to protect children online. This has led to new AI safety technologies. Companies like Yoti use AI to verify ages. HMD Global launched a smartphone that uses AI to block inappropriate content. These measures aim to keep kids safe from harmful material online.

How much is AI worth to the economy

AI is already boosting the economy through spending on hardware and infrastructure. Investment in AI has increased GDP. Companies are also preparing for AI to impact their operations, with more job listings asking for AI skills. However, some worry that too much money is being wasted on AI. It may take time for AI to truly transform the economy.

Podcast discusses AI's role in space missions

The "This Week In Space" podcast discusses how AI could be used on long space missions. Dr. Daniel Selva talks about a study on how astronauts interact with AI. The study looked at how crew members trust AI in the HERA habitat simulator at NASA. The podcast also mentions a model Falcon 9 rocket that people can launch.

AI browser can be tricked into scams and data leaks

Perplexity's AI browser, Comet, has security flaws that can be exploited. Researchers found that Comet can be tricked into buying fake products and leaking personal data. The AI assistant can't always tell the difference between safe and malicious content. This means it might follow bad instructions and give away sensitive information. Experts warn that AI browsers need better security to protect users.

Sources

Taco Bell AI drive-through AI ordering system Customer complaints System errors Dane Mathews McDonald's Wendy's TRM Labs Codex Vulnerability Agent Software flaws Security OpenAI Reinforcement learning AI agents Security risks API keys Guillermo del Toro Frankenstein Netflix Syngenta Soil health Farmers Climate adaptation GitHub Copilot Coding AI models GPT-4.1 Claude Opus 4 Debugging Security problems iRobo AI security robot India Peregrine Singapore Tenon Group Online safety AI safety technologies Yoti Age verification HMD Global Inappropriate content AI and economy GDP Job listings Space missions NASA AI browser Security flaws Data leaks Perplexity Comet