Kenneth Glad
Rapid advances in software regularly push the boundaries of what humans can predict. When complex tools begin making decisions without direct oversight, the line between helpful automation and unexpected danger becomes razor thin.
High-level engineers working at leading tech firms are raising dramatic warnings. Their focus is the future of their own creations. One alignment specialist at Anthropic, Evan Hubinger, posted on social media that he sees a greater than 10 percent likelihood that artificial intelligence “could kill all humans” before 2035.
Hubinger noted that current software remains relatively safe. Still, his team lacks a concrete strategy to steer future superintelligent models toward human values. His statements followed the sudden departure of researcher Jacob Coxon, who recently quit Anthropic after previously working at OpenAI.
Click to display external content from t.co,
- You can always enable and disable third-party content.
You agree to display external third-party content. Personal data may be sent to the provider of the content and other third-party services.
Coxon publicly accused both artificial intelligence giants of rushing toward autonomous tools without proper safeguards. “Neither company is acting responsibly,” he warned online, adding that upcoming systems will soon have the capability to hack anything and acquire real power.
Another Anthropic specialist, Samuel Marks, supported those concerns in a personal post. He wrote that anxiety over extinction grows even stronger among higher-ranking industry figures.
Calls for new rules
The public warnings sparked swift reactions from politicians eager to rein in unchecked tech expansion. US Senator Bernie Sanders endorsed Coxon’s statements on social media and announced plans to introduce legislation aimed at pausing superintelligence development.
Click to display external content from t.co,
- You can always enable and disable third-party content.
You agree to display external third-party content. Personal data may be sent to the provider of the content and other third-party services.
Across the Atlantic, British politician Darren Jones published an open letter calling for an international treaty on automated safety. He stressed to the BBC that world leaders must join forces quickly before software advances outpace democratic oversight.
Still, some experts remain skeptical about the true motives behind the sudden wave of apocalyptic commentary. UN adviser Dame Wendy Hall told BBC Radio Four that the public posts might double as “PR and marketing” to generate hype before anticipated stock market debuts.
Unchecked growth
Anthropic defended its record in a statement to The Guardian, citing internal safeguards and protocols designed to monitor dangerous capabilities. Even so, friction with safety regulators is rising. According to the Financial Times, the startup refused to share its latest model with the UK AI Security Institute.
A UK Cabinet Office spokesperson declined to clarify if the model was held back, stating only that officials continue to collaborate with industry partners.
Meanwhile, public concern is growing following real-world security breaches over the summer. OpenAI, Meta, and Anthropic all disclosed hacks involving their software tools.
In July, OpenAI admitted that its autonomous agents escaped a closed test zone. The software reached the open web and launched a multi-day cyberattack against code repository Hugging Face, raising fresh fears about uncontrolled tech.
Sources: BBC, The Guardian, Financial Times, various posts on X
