The makers of artificial intelligence (AI) models admit they only understand a tiny fraction of what goes on inside their models.
In January 2026, Dario Amodei, one of the founders of the AI company Anthropic — which created the AI model Claude — wrote an essay called “The Adolescence of Technology: Confronting and Overcoming the Risks of Powerful AI.“
“AI brings risks,” Amodei admits, “and because it is such a powerful technology, these risks are serious. They include the risk of losing control of AI systems, misuse of AI for cyber attacks, bioterrorism, and serious economic disruption.”
As Amodei sees it, “Humanity is about to be handed almost unimaginable power, and it is deeply unclear whether our social, political, and technological systems possess the maturity to wield it.” Amodei insists we must “map out the risks that we are about to face and try to begin making a battle plan to defeat them.” We vacillate between AI fears, which he calls “doomerism,” and AI opportunities.
“We are considerably closer to real danger in 2026,” Amodei warns, “than we were in 2023.”
Powerful AI, according to Amodei, “is smarter than a Nobel Prize winner … it can prove unsolved mathematical theorems, write extremely good novels, write difficult codebases from scratch. It has all the interfaces available to a human, including internet access. It can engage in any actions… taking or giving directions to humans, directing experiments, making videos. It does all of these tasks with, again, a skill exceeding that of the most capable humans in the world. It can solve tasks in hours or weeks, and then goes off and does those tasks autonomously, in the way a smart employee would.”
“It has no physical body, but it can control physical tools, robots, or lab equipment… It could even design robots or equipment for itself to use.” It can process information and act at “10 to 100 times human speed.” But “AI systems are unpredictable and difficult to control… many things can go wrong. AI models develop a wide range of undesired or strange behaviors, for a wide variety of reasons.”
Amodei outlines the concerns that AI’s impressive skills bring:
- Is this intelligence hostile, or does it share our values? Could it militarily dominate the world through superior weapons and cyber operations? A genius in everyone’s pocket could… essentially make everyone a PhD virologist who can design, synthesize, and release a biological weapon. Some biological agents could cause millions of deaths.
- Could a dictator gain dominant power, upsetting the existing balance of power?
- Could rogue actors (such as terrorists) use AI for “amplifying the scale of destruction?”
“This is a dangerous situation — the single most serious national security threat we’ve faced in a century, possibly ever,” he says.
To lower risks, Amodei suggests we embed third-party evaluators in AI companies to verify adherence to safety practices, pass national transparency legislation, require pre-release testing of models, require character and values training for the model personae, stop exporting chips to China; slow down the pace of model development, get AI companies within democratic countries to coordinate and establish common safety standards and limits on the rate of unchecked AI progress, and have U.S. and other democratic governments attempt to coordinate with authoritarian governments, taking seriously the challenges of verifying compliance.
“Current AI models are an almost endless gold mine of insight into both how to build AI well and what can sometimes go wrong with it if it isn’t built well,” Amodei says. “Society must have a say in how this technology is used, and more time for the necessary public deliberations.”
Anthropic has trained its Claude model to think of itself as an “ethical but balanced and thoughtful person.” Training an AI model is “not unlike with humans, who are raised with a set of fundamental values… but in any human there is some probability that something goes wrong.” An AI model may think “they are playing a video game and that the goal of the video game is to defeat all other players (i.e., exterminate humanity).”
Amodei believes AI will “raise the quality of life for everyone… contribute to enormous advances in biology, neuroscience, economic development, global peace, and work and meaning.”
In 1818, Mary Shelley’s “Frankenstein; Or, The Modern Prometheus,” also believed that “a new species would bless me as its creator and source; many happy and excellent natures would owe their being to me.”
“Humanity needs to wake up,” Amodei cautions. His essay “is an attempt — a possibly futile one, but it’s worth trying — to jolt people awake.”
Al Norman’s Pushback column is customarily published by the Recorder every first and third Wednesday of the month.