Former AI Employees Warn NYC Council About Out-of-Control AI

Former employees from Anthropic, OpenAI and Google DeepMind testified before the New York City Council on Monday, warning that AI companies are rushing to build powerful technology they cannot control. Jacob Coxon, a former Anthropic engineer, said humanity is more likely than not to lose control of advanced AI. The speakers described a reckless culture at major tech companies that prioritize speed over safety, and urged local regulators to act while Congress remains inactive.

Alex Turner, who left Google DeepMind, put the chance of an AI takeover at roughly one in three. He said misaligned AI is everyone's adversary. Kokotajlo added that the ability to spot misaligned AI is poor and getting worse. Representatives from OpenAI, Anthropic, Google and Meta told officials their systems are improving daily life and that they focus on safety.

Microsoft's Digital Defense Report 2026 says AI is shortening cyberattacks from days to minutes. Attackers use AI for hacking, phishing, and making malware. Government agencies are the most targeted sector at 27 percent, and the US faces the highest number of attacks. Defenders need AI tools to keep pace with adversaries who are leveraging automation at scale.

Chinese open-weight AI models from DeepSeek, Alibaba, Z.ai, and Moonshot are competing with Western systems at much lower prices. These models use techniques like efficient attention and quantization to run faster and cheaper. The AI market is splitting into tiers based on different needs, though cheap models can still be expensive to run on large hardware setups.

Mastercard and Rezolve Ai signed a worldwide reseller agreement. Mastercard can now market and distribute Rezolve Ai's AI-powered commerce tools to its merchants and banks. The partnership aims to help businesses create more personalized customer experiences and make AI commerce tools available to businesses of all sizes.

Trend Micro announced on October 5 that it is opening closed preview access to Autonomous Vulnerability Discovery in Trend Vision One. The AI-driven security service identifies and validates vulnerabilities in customer code and verifies that fixes have been applied correctly. Meanwhile, the preprint database arXiv is being overwhelmed by a flood of AI-generated papers, with submissions jumping from 20,569 in September 2024 to 26,646 a year later. arXiv has started limiting submissions to protect its quality.

In a lighter development, Cru Uncorked held a dinner where AI and human sommeliers paired wine with a four-course meal. The humans won by just one vote out of 19 guests per course. The sommeliers' knowledge of their guests helped them choose a sweet Sauternes for cheese that AI did not recommend.

Key Takeaways

Former Anthropic, OpenAI, and Google DeepMind employees testified at NYC Council, warning AI companies are rushing development they cannot control
Jacob Coxon said humanity is more likely than not to lose control of advanced AI; Alex Turner estimated AI takeover chance at roughly one in three
Microsoft's Digital Defense Report 2026 says AI has shortened cyberattacks from days to minutes, with government agencies targeted at 27 percent
Chinese models from DeepSeek, Alibaba, Z.ai, and Moonshot compete with Western systems at lower prices using efficient attention and quantization
Mastercard signed a global reseller deal with Rezolve Ai to distribute AI-powered commerce tools to merchants and banks
Trend Micro opened closed preview of Autonomous Vulnerability Discovery, an AI-driven security service in Trend Vision One
arXiv submissions jumped from 20,569 to 26,646 in one year due to AI-generated papers, prompting the database to limit submissions
Human sommeliers beat AI by one vote in a wine pairing challenge at Cru Uncorked's dinner

AI insiders warn NYC council about dangerous technology

Former employees from Anthropic, OpenAI and Google DeepMind testified before the New York City Council on Monday. They warned that AI companies are rushing to build powerful technology they cannot control. Jacob Coxon, a former Anthropic engineer, said humanity is more likely than not to lose control of advanced AI. State and city governments are considering new regulations while the federal government has taken a passive approach.

AI experts urge NYC leaders to oversee dangerous technology

Artificial intelligence insiders warned New York City leaders on Monday about the risks of emerging AI technology. They called for greater oversight as local governments respond to concerns about safety. The speakers described a reckless culture at major tech companies that prioritize speed over safety. They urged regulators to act while Congress remains inactive and the Trump administration has dismissed AI fears.

Former AI engineers testify at NYC Council hearing on safety

At a New York City Council hearing on Monday, former employees of Anthropic, OpenAI and Google DeepMind described dangerous workplace cultures at big tech firms. Jacob Coxon said the startup mindset of moving fast and fixing later does not work for the most powerful technology ever built. He warned that without transparency and safeguards, AI could spiral out of control. Representatives from OpenAI, Anthropic, Google and Meta told officials their systems are improving daily life and that they focus on safety.

Former AI researchers say industry risks human extinction

Jacob Coxon, a former OpenAI and Anthropic researcher, told the New York City Council that humanity is more likely than not to lose control of advanced AI. He said the AI industry is at elevated risk because companies use a move fast and break things attitude. Kokotajlo said the ability to spot misaligned AI is poor and getting worse. Alex Turner, who left Google DeepMind, put the chance of an AI takeover at roughly one in three and said misaligned AI is everyone's adversary.

Trend Micro previews AI-powered vulnerability discovery tool

AI vs Sommelier Wine Pairing Ends in Near Tie

Cru Uncorked held a dinner where AI and human sommeliers paired wine with a four-course meal. Guests tasted both and voted for their favorite pairing. The humans won by just one vote out of 19 guests per course. The sommeliers' knowledge of their guests helped them choose a sweet Sauternes for cheese that AI did not recommend.

Mathematicians Warn arXiv May Not Survive AI Paper Flood

The preprint database arXiv is being overwhelmed by a flood of AI-generated papers. Submissions jumped from 20,569 in September 2024 to 26,646 a year later. Experts say these low-quality papers are hard to understand and threaten the database's usefulness. arXiv has started limiting submissions to protect its quality.

Mastercard and Rezolve Ai Sign Global Reseller Deal

Rezolve Ai signed a worldwide reseller agreement with Mastercard. Mastercard can now market and distribute Rezolve Ai's AI-powered commerce tools to its merchants and banks. The partnership aims to help businesses create more personalized customer experiences. It marks a major step in making AI commerce tools available to businesses of all sizes.

Dan Kagan-Kans Joins as New Writer

A new writer named Dan Kagan-Kans has joined the team. He spent ten years as managing editor of Mosaic, a magazine about Jewish history and culture. He also writes arts criticism and essays. He recently had a son and started reading about AI two years ago, which sparked his interest in the field.

China Faces Challenges Funding AI Safety Efforts

There are differing views on how China approaches AI safety. Some believe China is less worried about AI risk because its people tend to be optimistic about technology. Others argue China is now taking safety more seriously after leaders like Xi Jinping gave speeches about controlling AI risks at major events.

Microsoft report says AI speeds up cyberattacks

Microsoft's Digital Defense Report 2026 says AI is shortening cyberattacks from days to minutes. Attackers use AI for hacking, phishing, and making malware. Government agencies are the most targeted sector at 27 percent. The US faces the highest number of attacks. Defenders need AI tools to keep up.

Wake Tech hosts first AI Innovation Week

Wake Tech held its first AI Innovation Week with over 200 attendees. Students, faculty, and staff joined workshops and panels about AI in learning and work. Professors Dr. Kim and Dr. Patel discussed how AI can improve education and healthcare. The event aimed to explore real-world uses of AI and its future impact.

Chinese AI models change global industry economics

Chinese open-weight AI models from DeepSeek, Alibaba, Z.ai, and Moonshot are competing with Western systems at much lower prices. These models use techniques like efficient attention and quantization to run faster and cheaper. The AI market is splitting into tiers based on different needs. However, cheap models can still be expensive to run on large hardware setups.