OpenAI, Anthropic AI Safety Concerns Raise Alarm

Key Takeaways

• Former OpenAI researcher Jacob Coxon quit the industry, warning that Anthropic and OpenAI are racing toward self-improving superintelligence without enough safety measures.
• Coxon stated that the current approach to AI development puts human lives at risk and called for more thoughtful research practices.
• Visa CEO Ryan McInerney reported that 75% of consumers do not trust AI agents to handle payments, though trust exceeds 70% with Visa involvement.
• McInerney noted this data was shared during a conference held on September 8, 2026.
• North Carolina reported $15.6 million in losses from AI-related internet crimes in 2025, the first year this category was tracked separately.
• Nationwide, the FBI received over 22,000 complaints about AI crimes, leading to more than $890 million in total victim losses.
• Common AI-facilitated scams include investment fraud, business email compromise, and romance scams using fake profiles or voices.
• OpenAI is developing robotic researchers designed to perform complex tasks in programming, science, and medicine with less human intervention.
• OpenAI emphasizes that human oversight remains crucial for accuracy, especially in sensitive areas like medicine.
• NVIDIA Dynamo can accelerate multimodal AI model serving by up to five times using a new Encode-Prefill-Decode disaggregation technique.

Former AI Researchers Warn Tech Companies Are Gambling With Lives

Former and current researchers from Anthropic and OpenAI are warning that the tech industry is moving too fast with artificial intelligence. Jacob Coxon, a former OpenAI researcher, stated that both companies are racing toward self-improving superintelligence without enough safety measures. He believes this approach puts human lives at risk and has called on other researchers to be more thoughtful about the future. Coxon recently left the industry to highlight these serious concerns about the direction of AI development.

AI Researcher Quits Claiming Tech Giants Are Gambling With Lives

An Anthropic researcher named Jacob Coxon quit his job because he believes AI labs are taking dangerous risks. He stated on social media that neither OpenAI nor Anthropic is acting responsibly in their rush to develop advanced artificial intelligence. Coxon warned that these companies are racing straight toward self-improving superintelligence and are essentially gambling with human lives. His departure highlights growing fears within the scientific community about the speed and safety of current AI development.

Claude Creator Warns AI Race Is Out of Control and Dangerous

Jacob Coxon, a researcher who worked at Anthropic and OpenAI, has announced he is leaving the AI industry. He cited serious concerns that both major companies are not acting responsibly enough regarding the risks of advanced AI. Coxon explained that his experience at these top firms convinced him they are racing directly toward self-improving superintelligence. He believes this rapid development puts humanity in danger and has decided to step away from the field.

Visa CEO Says AI Shopping Is Here But Autonomous Payments Are Not

Visa CEO Ryan McInerney stated that consumers are using AI to shop but are not yet trusting AI agents to handle payments. During a conference on September 8, 2026, he explained that while people use large language models to find products, they still go to websites to complete transactions. McInerney said the main barrier to autonomous payments is trust, as three-quarters of consumers do not trust AI agents with their money. He noted that trust increases to over 70% if Visa is involved in the payment process.

OpenAI Developing Robotic Researchers for Science and Medicine

OpenAI is creating a new generation of artificial intelligence systems designed to perform complex research tasks with greater autonomy. These robotic researchers can understand problems, gather data, and propose solutions in fields like programming, science, and medicine. Unlike traditional tools, these systems can execute multi-step tasks and work continuously toward specific goals with less human intervention. However, OpenAI emphasizes that human oversight remains crucial to ensure accuracy, especially in sensitive areas like medicine.

Anthropic Insiders Warn Advanced AI Could Kill All Humans

Insiders from Anthropic have warned that the risks of advanced artificial intelligence are real and growing rapidly. One anonymous insider stated that the technology is reaching a point where humans may not be able to control it. The warnings highlight fears that AI could be used for malicious purposes to harm people, not just cause accidental damage. These concerns come as other leaders, including OpenAI CEO Sam Altman, have also expressed worry about AI speeding into an uncontrollable phase.

Club for Growth President Discusses AI and a 32-Hour Workweek

Club for Growth President David McIntosh joined a show to discuss if artificial intelligence could make a 32-hour workweek possible. He addressed questions about whether AI advances will change how much time people spend working each week. McIntosh also responded to political concerns regarding Republicans maintaining control of the House and the stakes in the Texas Senate race. The discussion covered both the potential economic impact of AI and current political events.

Match CEO Says AI Chatbots Could Worsen the Loneliness Epidemic

Match Group CEO Spencer Rascoff believes that relationships formed through AI chatbots could make the loneliness epidemic worse. He argued that while AI can help people connect, it might also create unrealistic expectations and make users feel more isolated. Rascoff emphasized the importance of real human interaction and urged people to take an active role in forming meaningful relationships. His comments come as the dating industry watches how technology continues to change human connections.

FBI Tracks AI Crimes as North Carolina Loses Millions to Scams

The FBI has started publicly identifying and tracking crimes committed using artificial intelligence as a tool. North Carolina reported $15.6 million in losses from AI-related internet crimes in 2025, marking the first year this category was tracked separately. Nationwide, the FBI received over 22,000 complaints about AI crimes, resulting in more than $890 million in victim losses. Common scams include investment fraud, business email compromise, and romance scams that use AI to create fake profiles or voices.

NVIDIA Dynamo Helps Speed Up Multimodal AI Model Serving

A new technique called Encode-Prefill-Decode disaggregation can accelerate the serving of multimodal AI models by up to five times. This method separates the vision encoder stage from the prefill and decode stages to improve efficiency. It works best for image-heavy prompts and short-to-medium outputs, especially when used with quantized mixture-of-experts models. Developers can use NVIDIA Dynamo to implement this technique and improve the performance of their applications.

Sources

• Anthropic researchers warn artificial intelligence could destroy humanity in the future
• Anthropic researcher quit, says AI labs are 'gambling with our lives'
• Claude researcher warns 'AI race is out of control'
• Visa CEO Says AI Shopping Has Arrived but Agentic Payments Haven’t
• The robotic researcher... OpenAI is pushing artificial intelligence from programming to discoveries in medicine and science
• Anthropic insiders warn AI could kill all humans
• Could AI make a 32-hour workweek possible? David McIntosh weighs in
• Match CEO: AI chatbot relationships could worsen the loneliness epidemic
• FBI now publicly identifies, tracks AI used in crimes. North Carolinians have lost millions
• When to Use Encode-Prefill-Decode Disaggregation to Accelerate Multimodal Model Serving