Anthropic researcher quits, warns AI race could threaten humanity

ASM, Philippine Canadian Inquirer
September 13, 2026

A former Anthropic researcher has resigned over concerns that leading artificial intelligence companies are moving too quickly toward increasingly powerful systems without adequate safeguards.

Jacob Coxon, who previously worked at both Anthropic and OpenAI, said the race to develop artificial superintelligence could carry catastrophic risks if companies lose the ability to control systems more capable than humans.

Coxon said researchers inside major AI laboratories genuinely worry that advanced AI could pose an existential threat, accusing companies of effectively “gambling” with humanity’s future as they compete to reach the technology first.

His warning was echoed by Evan Hubinger, Anthropic’s alignment science lead, who said he personally puts the chance of AI causing human extinction within the next decade at more than 10%.

Hubinger also said researchers have not yet solved the problem of ensuring that a future superintelligent AI reliably follows human intentions and remains under human control.

The estimates are the researchers’ own assessments, not established scientific forecasts. There is no consensus that current AI systems are close to wiping out humanity, and today’s AI models are not considered artificial superintelligence.

The concern instead centers on what could happen if future systems become substantially more capable, autonomous and able to improve their own performance faster than humans can monitor them.

What is AI alignment?

In AI research, alignment refers to efforts to make sure advanced systems behave according to human goals and safety requirements, even as their capabilities increase.

Anthropic itself describes aligning superintelligent AI as an unsolved problem. Its researchers have studied scenarios in which models display behaviors such as deception, attempts to avoid shutdown, and other forms of “agentic misalignment” in controlled evaluations.

The company stresses that current model capabilities have not reached the level where such behaviors would pose catastrophic risks, but says stronger safeguards will be needed as AI becomes more powerful.

Recent Anthropic research has also examined real-world cybersecurity incidents and the growing ability of AI systems to assist with intelligence and weapons-related tasks, underscoring concerns about how more capable systems could be misused.

Calls to slow down and regulate AI

The debate is no longer confined to AI laboratories.

In the United Kingdom, Labour MP Alex Sobel introduced an Artificial Superintelligence Bill on Sept. 8 that seeks to prohibit the development, deployment and operation of artificial superintelligence systems. The proposal is still at an early stage in Parliament and has not become law.

OpenAI Chief Scientist Jakub Pachocki has also called for greater coordination as AI capabilities advance.

In an essay published this month, Pachocki said no AI laboratory has yet solved alignment and monitoring well enough to continue scaling indefinitely at maximum speed. He said voluntary slowdowns may eventually be necessary until companies agree on common safety standards and urged governments to make international coordination a priority.

OpenAI has separately backed mandatory national AI safety requirements in the United States, arguing that companies and governments need stronger rules as frontier models become more capable.

The warnings highlight a growing divide over how quickly advanced AI should be developed.

Supporters of rapid progress point to potential benefits in science, medicine, education and productivity. AI safety advocates, meanwhile, argue that the possibility of systems becoming difficult to control warrants stronger oversight before capabilities advance further.

For now, claims that AI could cause human extinction remain predictions rather than established outcomes. But the fact that researchers working directly on frontier systems are publicly raising the possibility is adding pressure on governments and AI companies to answer a central question: how far should AI development go before stronger safeguards are in place?