The artificial intelligence industry is experiencing one of its most heated debates yet over whether the technology could pose an existential threat to humanity. During a recent episode of TechCrunch’s Equity podcast, hosts Kirsten Korosec, Sean O’Kane, and Anthony Ha discussed the latest alarming warnings from AI researchers and executives. Anthony Ha expressed skepticism about many of the doomsday predictions, while Kirsten Korosec questioned whether these statements were simply a way to show off how advanced their companies' AI models are. The conversation was edited for length and clarity, but it highlighted the growing tension in the field. One of the warnings came from Jacob Coxon, an Anthropic researcher who resigned after working on pretraining research at Anthropic and OpenAI. Coxon said leading AI companies were gambling with people’s lives by racing toward systems that could improve themselves. Anthropic alignment researcher Evan Hubinger then said he believes the chance that AI could kill all humans is greater than 10% within the next decade.
Sean O’Kane described how quickly recent developments have unfolded, calling it a strange and intense moment in AI history, according to TechCrunch. He noted that the timing of these warnings came after a series of high-profile events, including a hack at Hugging Face and new model releases from Anthropic and OpenAI. Anthony Ha disagreed with O’Kane’s view, stating that if AI could really destroy all of humanity, then such warnings deserve strong emphasis. Ha said he remains skeptical of doomsday narratives even though he believes possible catastrophic risks deserve attention.
The discussion also touched on how difficult it is to speak about the AI community as a whole, since there are many different perspectives and individuals involved. Anthony Ha pointed out that some numbers like a 10% chance of AI causing human extinction are often made up without solid data behind them. She questioned whether the warnings were just a way to highlight how powerful their own AI systems had become. Anthony Ha admitted he had wondered about this too, but he didn’t think it was entirely cynical or a marketing trick. He believed that many people in the field do have real concerns about the direction AI is heading. However, he also noted a personal tendency among researchers to see their work as the most important and dangerous thing in the world. Sean O’Kane said recent incidents at OpenAI gave him the feeling that some companies do not fully have a handle on the systems they are developing.
After the podcast was recorded, Anthropic CEO Dario Amodei called for the industry to slow the pace of advanced AI development. Amodei proposed giving outside evaluators broad access to company systems so they can check whether safety measures are being followed. His proposal added another major voice to a debate that has moved beyond researchers and into company leadership.
Separately, TechCrunch reported on the latest Y Combinator Demo Day, which featured a batch of startups focused heavily on deep technology. This group of companies stood out for their focus on advanced tech solutions like nuclear-powered data centers and energy-efficient AI hardware. Investors were impressed by the innovation but also noted that valuations in this batch were more realistic than in recent years. TechCrunch said the Summer 2026 group leaned much more heavily toward deep technology than several previous YC batches.
One standout company was Atomarine, which plans to build floating nuclear-powered data centers at sea. The company has already secured over $4 billion in customer interest through letters of intent. TechCrunch described the $4 billion figure as customer interest expressed through letters of intent. Atomarine says it plans to launch a gas-powered pilot by 2028 and later move to floating nuclear power ships. Y Combinator says the company plans to place standardized data center barges offshore where seawater can provide cooling.
Another company, Dipole Labs, is developing high-speed optical networking hardware designed to reduce energy use in AI data centers. Dipole Labs says its optical switch keeps data in the optical domain instead of repeatedly converting it between light and electricity. The company says this could reduce power use and heat while allowing expensive computing hardware to spend less time waiting for data.
Lamb Labs is working on custom chips that hardcode AI model weights into silicon to improve efficiency. The company calls the planned hardware a Model Processing Unit. Lamb Labs says its approach is meant to reduce the memory-bandwidth problem faced by conventional AI processors.
Praxis AI collects real-world data to train robots, partnering with businesses to gather video and training material. Praxis AI says it has already collected training data in more than 150 real-world environments. TechCrunch reported that the company is already working with publicly traded businesses.
Nori is a company building affordable humanoid robots for everyday tasks, with early sales already reaching nearly half a million dollars. Nori says its A3 robot will cost $1,688 and is expected to ship in fall 2026. The robot is being designed for jobs such as loading dishes, folding clothes, fetching items and helping in the kitchen.
Cosmic Robotics is focused on autonomous robots that can lift heavy objects, with plans to use this tech in Mars colonization efforts. TechCrunch reported that Cosmic Robotics says it already has $25 million in contracts through 2027 and is using its technology to install solar panels. The founders eventually hope similar robotic systems could help automate construction on Mars.
Parasma is exploring how human brain cells might one day be used for computing tasks. TechCrunch reported that the company is studying whether biological computing could offer a more energy-efficient alternative to conventional AI hardware.
Waddle Labs aims to create an API that writes robot control code, hoping to bring a ChatGPT moment to robotics. Waddle Labs says developers will be able to connect hardware to its API and give robots instructions in natural language. Its agents would then generate the control code needed for the robot to carry out the request.
These companies show how the AI field is expanding beyond just software into hardware and robotics.