Anthropic CEO Dario Amodei has published a new essay calling on the AI industry to deliberately slow the pace of capability advancement, warning that unchecked progress could let AI systems seize control of major parts of the internet within roughly six to twelve months.
AI Doomsday Soon? #
In the essay, titled "We Must Pace the Frontier," Amodei remarks that two developments over the past several months convinced him that safety work is falling behind capability gains.
- He points first to recursive self-improvement, AI systems increasingly used to build the next generation of AI. He says this is accelerating faster than expected across the industry.
- His second concern centers around the OpenAI and Hugging Face incident, in which a swarm of AI agents conducted unauthorized cyberattacks and worked collectively to secure the group's success. Later the swarm of bots tried to hack the very system meant to evaluate their performance.
Given the accelerating rate of AI capability development, it’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage), and that the scale of damage would continue to increase from there if AI becomes more powerful without the necessary guardrails.
Amodei writes that a more capable version of that same swarm type could realistically build a persistent botnet hacking much of the internet. He believes this will cause hundreds of billions of dollars in damage.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our…
September 12, 2026 We have tried to prioritize caution over speed and prudence over profit.
Pacing The Frontier #
To address the doomsday scenario, Amodei proposed a three-part plan in the essay he calls "pacing the frontier."
- The first step will involve evaluators with employee-level access to monitor safety practices and give third-party reviewers desks, badges, and the ability to publish findings without company editorial control.
- The second step calls for coordination among AI companies in democratic countries on shared safety standards.
- The third step seeks cooperation with authoritarian governments, including China, though Amodei calls that outcome far less certain.
Will this work?
We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.
Read Daroi’s essay here. See Also: Ex-DeepMind Scientist Alex Turner Reveals His Job At Google Was To Plan Failsafe For Killer AIs
[Tech](/tech),
[AI](/ai-1),
[Doomsday](/doomsday),
[Anthropic](/anthropic),
[Artificial General Intelligence](/artificial-general-intelligence),
[ai agents](/ai-agents),
[Dario Amodei](/dario-amodei),
[AI apocalypse](/ai-apocalypse)